Reworked the envelope to have a linear attack and with gentler control of the decay/release times. The linear attack was a Moog thing but was not universally appreciated - it does not work very well for long attack times. Several synths such as the Synthi preferred the use of exponential for this and other reasons. The actual envelope code does not have linear or exponential attack as a fixed value, it can be configured by the emulator although at the moment they all just take the default setting, now linear. There is a ./configure option to have them default to exponential rather than linear: --enable-exp-attack
It is now also possible to build with an alternative ADSR max ramp duration, the following will give you a maximum ramp of 30 seconds, default is 10s.
./configure BRISTOL_RAMP_RATE=30
These changes were to correct an anomaly in the previous design where the actual ramp was a function of the sampling rate, hence if you step up to 96KHz or even 192KHz for the improved sound quality then the ramp times would change (now fixed). If you want to use very long attack times then exponential attack will give arguably better results, that is admittedly opinion.
NOTE that these envelope changes will affect existing patches to some extent.
Added code to have fanning linear pots. The further the mouse is from the axis of the pot the finer the value changes become.
Resolved a error in midinote.c that broke the scala file support. A patch was posted to SFDN a long while ago now, this just rolls it into a release.
MIDI Program Change events were not being filtered by channel - they were being tracked as OMNI. Applied same change to Bank Select. Have not tested all the multilayered synths, they have exceptions since some use separate memory indeces and some use combined settings.
Reconfiguring the continuous controllers as things like inverse mappings were failing. The issue was that both the engine and GUI might be using the same mapping file, dual inversion negated the effect. Corrected this with only doing the mapping in the GUI but added a token called modWheel for the 'mcm' file to allow the engine to have alternative curves for the modulation wheel.
The request to have an inverted modWheel requires the following: Start the engine with -tracking to prevent the GUI tracking some events. Edit mini.mcm and add in
modWheel: 127 0.001
modWheel: 0 127.0
This file is emulator global, ie, to have different settings with the same synth then a user would have to configure different bristol cache variables. A better way to do this might be to have an ALSA event manager in front of the bristol emulators.
Ported over some note loss fixes from the 0.70 (smartphone) stream where with excessive system load there were lost notes. This was never actually reproduced on Linux so will discuss the fix with end users. The original change resolved the issue when seen on a 600MHz ARM A8 on Maemo/N900 from Nokia and which were also tested/confirmed on Android. It concerned signalling between the note allocation code and the ADSR envelope states.
The multilayered synths would lose tracking of the PitchWheel - the pitch was correctly calculated but only applied to the first bristolAudio structure in the chain.
The Polysix gives click/tick at high frequencies and mods. The cause was the change from the precached buffers to recalculated buffers. The code was changed to have the buffers always calculated/cached on note events as a fix to the existing band limited oscillator code. The code change was only the prophetdco but it could be pushed into the other oscillators.
The BME noise mix was all but silent, readjusted the gain settings for the mix.
The Pro1 and Prophet10 saved sequencer memory was not reloaded on restart. Had to apply changes to a few different parts of the arpeggiator/sequencer code.
Gain levels seemed low on the ARP 2600 and the Direct Out final stage gain left a signal that was almost inaudible.
Added the option to use a memory barrier over the jack ringbuffer. The issue is that the JRB can be demonstrated to fail on ARM CPU due to them having weak memory ordering. Intel should not be affected. AMD is a marginal case. The config option is --enable-memory-barrier which defaults to no barrier since most of the installs will probably be on Intel.
Packaged a version of bristol for the Nokia N900 running Maemo. As of this first release only the MiniMoog is operational however a larger selection of them are feasible for this handset. There is a rather large latency due to having to buffer a lot of data into pulseaudio. The code will work better if you step the CPU min up to 500 or 600 MHz with cpufreq-set or similar tool, it will not lower the latency but without this feature there can be quite a few underruns.
For the installation you need to download the file to somewhere on your MicroSD, it then needs to be gunzipped and untarred (which was the only way I could get the DEB onto the sf.net Download File page). Then you can do a normal install:
gunzip bristol-0.60.8.deb.tar.gz
tar xf bristol-0.60.8.deb.tar
dpkg --install bristol-0.60.8.deb
The result will be an icon in your applications folder which will stay blue until you restart your handset (am working on this to be automated). The code is optified to reside on what should just be the MicroSD card and memories are saved do the Documents folder for inclusion in backups of the handset. The unit should respond on the X/Y tilts for pitchbend and modwheel respectively.
The sound quality of the desktop integrated commands is not high, corners were cut to keep the CPU and battery footprint low.
The file can be downloaded here.
This release should be considered a beta for Maemo.
Made some changes to the way bristolrc is handled. If the file exists then a seach is made for two tokens, PREARGS and POSTARGS. If they don't exist then the same default behaviour takes place: all the contents are concatenated into a set of options that are inserted before all the user parameters. If either of these tokens are found then PREARGS are inserted before the user parameters and POSTARGS are appended to the whole net commandline to the GUI and engine. This gives a lot more flexibility to override default behaviour, for example:
PREARGS=-count 1024 -rate 48000
POSTARGS=-jackstats -jack -multi 0
Using such contents then the command 'startBristol -mini' would be expanded to
startBristol -count 1024 -rate 48000 -mini -jackstats -jack -multi 0
A part of the reason for the above change was that there were reports of issues with the ARP 2600 and period sizes less than 1024. The culprit appeared to be the -multi option, this defaulted to 4 for the 2600. Changed it to default to zero with a need to then explicitly reconfigure it. The reconfigure can now be automated in POSTARGS. The diverse Bristol front end appliations such as monoBristol provide user arguments to the startup script, the current bristolrc definitions can provide capabilities to extend how they operate as there are now pre and post arguments that encapsulate the frontend parameters.
This release is noted as including hammond optimisations, these are not with respect to CPU load, rather with respect to accuracy of the emulation:
Adjusted the B3 Preacher Compression code to provide a more pronounced non-linear amplification. The results are interesting/different. It mixes well with the new gearbox generated for a clean tonewheel (mostly sinewaves with no crosstalk) and a mild crosstalk gearbox with some sliced tonewheel harmonics. The control is not continuous, there are 3 stages which are selected as the rotary changes position, this would normally be represented with radiobuttons however the GUI previously only had a rotary control so this was used to prevent a big reordering of the options panel layout.
The new tonewheel gearbox are in the factory shadow profiles, per default as below. To test them then copy the target gearbox to your private memories (these are the defaults, your installation may be different):
source: /usr/local/share/bristol/memory/profiles/tonewheel*
dest: ~/.bristol/memory/profiles/tonewheel
Also made few alterations to the B3 VC algorithm changing the tap gains and LC filter values, altered the tap selections for the different 1/2/3 settings and adjusted the selector mode to reconfigure LC and phase (which work in combination) rather than the actual tapping points. More work could be done but these do now give some different characteristics where previously VC3 was the only useful setting. [Note: it would actually be quite easy to make even further improvements here by using an alternative filter algorithm into the different taps. The current one is a rooney which has weak phase modification, it is similar to the original in as much as that could only introduce about 1ms of phase change however bristol has other algorithms that would give better phase changes by frequency which would be closer to the Hammond vibra. The current code overcomes the rooney limitation by taking the phase changed filter signal (which has a phase change that is dependent on frequency) with an additional phase change that is time constant. The issue is efficiency, the rooneys are single pole and come almost for free. ffs]
Reintroduced some B3 code to give contact delays for each drawbar. The delay is configured in the tonewheel file and is then a function of velocity. The tonewheel file defines the maximum sample delay before any given harmonic is activated and this delay is reduced depending on velocity. It is also possible to configure a click sample per bus out of 4 available however the author will state that none of the samples is accurate: click, thump, shelf, etc, is not an accurate emulation. Depending on which source is considered authoritative then the Hammond click may be a pure function of tonewheel position and its activation point. This code emulates that effect with the option of adding in extra effect as a combination of the tonewheel file samples and the overall click setting. Note that the 'click' level does not affect the cut in points of the harmonics, reducing the 'click' level will not affect the tick sound when a drawbar contact is finally made. The harmonics do not have a constant phase (as per the original) however the tonewheel file can override that to emulate the L and M series spinets. With the current code, pressing a key very slowly should result in the delays between the different busses being almost audible, an effect that reduces with velocity. The configured Click is reduced when the vibrachorus is on as is also heard on the originals. This is partially due to parameterisation and partly due to the filtering effect of the vibra-chorus.
This click code has actually been available for a few years, it was removed as it affected other developments for gearbox optimisation. This optimisation was to 'post generate' the tonewheels: if a harmonic was not needed then it would never be generated. This was not condusive to busbar delays so now the decision has been taken to drop the CPU optimisation in preference of a more accurate rendition. At the same time the tonewheel damping was corrected so that, under control of the Damp option, the summed levels of any given wheel is not linear but is a function of the number of times it is tapped. It is noted here that if no click is desired, at all, then the 'preacher' algorithm should be disabled at which point notes are generated from zero point as per many of the commercial emulators. Several of the later Hammond models reduced/eliminated key click using various techniques, disabling the Preacher would give results closer to the B2XX series organs. Disabling the preacher removes the damping, busbar offsets, and all the other features of the tonewheel file.
Adjusted the click code to introduce changes to the bus offset delays by key as well as velocity. The existing code used a single profile per gearbox but no two keys would have had the same contact profile as this was based on aging of the springs: the code now has semi-random delay selection. The 'semi' refers to code that will select the same relative offsets if the same key is sounded sequentially, adjusted by velocity still: this is how Jimmy Smith used to play key click with the same key - now the delays for that sequential key do not change (they will if intermediate notes are sounded which he also did and potentially the delay offsets may be prebuilt for the whole keyboard rather than calculated, ffs). The different bus delays only affect timing not the actual effect so if 'Click' is non-zero then the sample selection per bus will be constant even though the delay changes. Now, another affect of the bus delays is that if playing quickly it is possible that a note will start and stop before even all the harmonics have sounded, this is another trick of Smith when he played notes to not traverse the complete travel of the key: not all harmonics would actually make contact and this is also possible with bristol (depending on velocity, buffer size and configured bus delays).
Added a -keytoggle flag, it affects GUI behaviour where a key is released when the mouse button is released. Previously the key state latched which was useful for polyphonic emulators but not necessarily for the monophonic ones. The default operation is now not to latch the keys, this option overrides the new default behaviour to give latching keys again.
There were some sticky keys and note 'ticks' when using the monophonic synths with -nnp (no note preference). The polyphonic note logic could completely kill notes when overvoice and if a keyoff event occurs for a voice on the new list then some of the flags would get damaged. None of this happens with monophonic note preference or with polyphonic synths (well, the ticks may still happen if the polyphony is exceeded but get burried in the mix of multiple voices, ffs).
Bug report on some emulators defaulting to only one semitone of pitchwheel depth. This turns out to be emulator aliases: the interface recognises some alias names but it could not set defaults except on exact names. Incorporated some changes to accept the known emulator aliases too.
Feature request that the SID should acknowledge pitchbend wheel as well as or instead of pitchbend on the modwheel. Not a big change and the issue with the MOD wheel is that it was difficult to tune having no notch. Currently both controls can affect pitch but the modwheel can be turned on/off selectively.
There was a window when the global configuration options would not be sent to the engine, a race condition. The result would be no output signal.
There was another race condition in the GUI library which for now has been patched. The actual fix is a little larger but it will be left for future research.
Found a bug in the MIDI handling code for NOTE_ON velocity zero being mapped to NOTE_OFF velocity 64. When the ringbuffer was introduced the check moved to the wrong side of the ringbuffer causing an endless loop at 100% CPU, for some emulators quite a lot of nasty random noise and potential system 'hang' for certain CPU configurations with RT scheduling. This had not been reported which means either people did not bother or the use of NOTE_ON velocity zero is not widely used.
Release 0.60.7 had an undocumented feature, undocumented because it was not that big but it did go into the manual page. The option -jackstats will prevent bristol from attempting to find out the current jack server settings, these parameters would instead be defined with -count and -rate. This change fixes some issues with jack connections, with LADI, and makes the GUI start a bit faster as it does not have to poll the server from the script then reconnect to the server from the audio thread.
The BME-700 Glide on/off switch was not active.
Roadrunner Vibrato was too deep, by a long way. Scaled back the tremelo at the same time. Gentle vibrato was still possible but not under mouse control, it had to be done with keyboard accel/deccel.
Added in an option called -memdump which will copy all memories for a given synth (specified with -emulate) to a target location. That location is then used as the session cache so further changes will be made to the dumped memories. That is intentional: this can be used with Session Management using a target location in the session directory so that the session will include all of the synth memories. The copy operation of the user memories only happens once when the directory is created.
Added in an option call -neutral, this will initialise the emulator with the value zero for all parameters. This provides a clean sheet from which to start sound generation. The code is generic and there are some exceptions to where it is applied: the organs and pianos do not implement the option, some of the synths are exceptions too (P800, BitOne, DX) as they have a non-intuitive interface that does not lend itself to being zeroed. For testing the few that do not implement the option then '-load -1' can be used as a potential alternative.
Added a runcom file "bristolrc" located in ${BRISTOL_CACHE}/bristolrc which may contain the parameters that are always provided on the command line. The parameters will be placed before the command line arguments so that they can be overridded but they provide the possibility to simplify the command line that needs to be specified when bristol is invoked. The variable BRISTOL_RC can override the default location and the -rc option will circumvent it.
The size of the voice table is now a configure time parameter. There were some requests for optimisation and reducing the size of these tables is a small one. ./configure BRISTOL_VOICECOUNT=32 is the default. This is not the max table size which continues to be 128 voices: the -voices option is a runtime parameter that can be used to increase the size of the table above the VOICECOUNT limit, this ./configure option is just the default.
Resolved an issue with the filter tuning of a few of the emulators, they were using an incorrect keyboard tracking value which meant that with the filter at high emphasis there was discordance in the oscillator vs filter waveform frequencies.
Reworked the summary text for small changes to some of the synth names: they are sometimes different from the switches used to start them however the actual name is required to be able to use emulate correctly. The -summary list will list the emulators that have been finished, --summary will list all of them even though a selection do not actually operate yet.
Removed the empty pop-up menus. The code is still there however it is no longer called by either GUI or library. There were diverse requests to stop posting them as they had no purpose.
Added an initialisation watchdog timer. If the audio thread does not activate in this period then the engine will gracefully exit rather than just hang around waiting for somebody to talk to him. The default timer is 30 seconds but can be changed with the '-watchdog [s]' parameter.
The ARP Axxe was about a semitone out of tune. The cause was the logic behind the pitch buttons which was broken.
Glide parameter settings could be lost on restart, the 'glidemax' parameter was not being correctly initialised so remained as zero.
Remedied a test failure for the jackd binary. If jack is requested as the audio interface then we need to connect to it to find the correct sampling rates and period sizes. The check for failure of this request was broken.
Fixed a very arbitrary race condition between the audio and midi threads causing segfaults which only happens during initialisation with heavily loaded CPU. Doubt if this will have been seen outside of development as the window was extremely small.
There was a spurious 'newline' debug message when active sense was detected from a keyboard. Bristol does not support active sensing (except from the GUI) and whilst these messages did not cause any issue with the synth operation they were a great annoyance for the command line interface users. The print statement was put under a debug flag.
Was requested to implement a -geom option which would minimally allow for window placement onto the screen. This seemed very reasonable so incorporated it. The option will also take width and height parameters such as 500x200+100+100 and where necessary break aspect ratio and configure antialiasing. If only one of width/height is given then aspect ratio is maintained, and options such as +100+100 will just affect window positioning. Options like -geom 640 will just set the pixel width, maintain aspect ratio and take default placement, possibly invoking antialiasing as well.
Randomised the TCP port numbers used to link the GUI to the bristol engine. There remained an issue with the Jack Session Manager and configurations that used multiple invocations of bristol: the TCP port identifiers would conflict causing one of them to fail on reload. I think the option '-host unix:1234' is a work around the issue as long as '1234' was always unique per bristol but this only works on localhost as it uses unix domain sockets rather than TCP domain. The longer term solution was to randomise the port numbers in the range 1024 to 16383, the transient service ports since otherwise there is still a window for error if the TCP ports are not closed down cleanly (although that is now finally a rare occurence). The port identifier can still be overridden and will be inherrited by the JSM interface.
Resolved a few issues with different device drivers. OSS was actually broken for a while due to incorrect flags and buffers and there were a few ALSA anomalies although they were inadvertently correct.
Was requested to add a manual page for bristoljackstats for part of the debian hurdles. Took the chance to also remove two ancillary binaries which generated notes but were only for QA testing, not general distribution.
Added an option called -sleep which stalls the init process for
Corrected some ./configure options so that they correctly check for jack and alsa header files rather than just the packages and libraries before including them in the build. This change included adjusting the default drivers so that OSS is configured when jack and alsa are not available.
Fixed an issue with the SID where double clicking on the memory Load button would turn on the soft chip debug code. This was a development feature that was not intended to be distributed.
The -mono noise mod was originally just the literal noise source, this is not a great mod and does not mimic the original which I think used a S&H circuit on the noise source clocked by Osc3. Reworked this into the emulator anyway, the sonic qualities are better. This should actually also be put into a few of the other emulators (Prophets too for example).
The monophonic note logic no longer respected -nnp due to incorrect flag management. This needed to be corrected as it was backing off to HNP which implied that the engine ran constantly.
Discarded a check in the ShmImage code so that the check for support in the server was dropped. Most servers which apparantly support the SHM structures seemed to be responding negatively to the capabilities request which caused backoff to the Pixmap interface. The current situation is that when compiled with SHM support brighton will attempt to use it. If the server really has issues with that then the -pixmap options will have to be requested manually.
Resolved a timing issue that can occur at startup, the event manager needed to wait for positive status on the control interface before proceeding. Only affected single core systems due to threading, and was naturally intermittent.
Failure to initialise the operating gain of the envelope when in some monophonic note conditions could result in apparant loss of signal until the note states were fully completed. This would manifest itself as either no output, low output or even as lost notes over MIDI channels.
Midi debug flags were not being parsed correctly in the engine. Should not have affected runtime as it was only a debug which is not usually active.
Changed the startBristol shell to request /bin/sh in place of /bin/bash.
Coded an interface into the X11 SHM extensions under the configure switch --enable-shmimage, some that was needed since the latest X11 servers have dropped their support for XImage interfaces. This left only Pixmap which works but is rather slow.
Added an option to restrict the rotary controller tracking to pure mouse up/down rather than geometric positioning. This was on request and is -rud or -rotaryUD, the value will be a fraction of the current window height. The code will respect notched controls and similar stepped rotary types.
There was a bug in the event forwarding code that would damage intermittent messages. The issue was masked for a few reasons: the key events did not use this code path, controller events would be generated in such high volumes that the loss was not visible. Program change was affected, also tacked on code to address bank change events which were previously not distributed to the engine although the way MIDI does bank management is not really consistent with bristol memory management: bristol tends to work with banks of 8 or 10 memories as per the old synths. MIDI runs with banks of 128 memories and the result will be some overlap. It works as long as a single access method is used.
Added -stop as a method to reap any existing bristol processes. It will kill the audio process if running which will result in any GUI dying due to the active sensing. All processes will be terminated which may not be the target action if multiple engine are being used, the options 'startBristol -kill -mini' will then only terminate the engine and GUI associated with the mini emulator. This is also not infallible as one engine could be running multiple emulators however the goal was really just to terminate a daemonified engine rather than do any really intelligent process management.
Added -cache as a runtime option to allow specification of the cache path for memories and profiles.
Reincorporated code for channel pressure recognition. It had been dropped from an earlier release due to dumping core when applying the channel change to all the voices associated with that channel. Also fixed up some issues with the poly pressure handling, I think it was want to hit 100% CPU. This was also ported back to 0.50.8.
Fixed a bug that had been reported a while ago against the MG1/Realistic of corruption of the GUI image and loss of sound. The GUI corruption was due to incorrect bounds checking on some bitmap manipulations which were corrected, a fix that is applicable to all the emulators. There were a couple of other specific issues with the MG1 and RhodesBass that had other consequences, also fixed. This was ported back to 0.50.8.
Added in another velocity map: no velocity sensitivity. That should have been there from the first implementation of maps but I think I got carried away with the ability to add hundreds of different curves that this got swept aside. It can now be requested with '-curve 0' to give something like a -3dB constant velocity level for all struck keys. Ported back.
Put the autozoom feature under a configure time option --disable-autozoom, some issues were reported where it would be apparantly get activated unexpectedly, ie, the windows would constantly flip in size even though the -autozoom was not given as a run time option. The ./configure option also removes the Enter/Leave tracking which was the more probable cause of the error. There may still be some issues with KDE4.4 that were reported. This code was tested against 4.3 and it does deliver a lot of seemingly superfluous ConfigureNotify which were give the window a Postage Stamp effect. Filtered out the ludicrously small window sizes that were being proposed. KDE 4.4 testing is pending.
Added a preinstalled version check to the build scripts to exit ./configure early if bristol is already on the system. The issue is that library directories may be different between the two, most notably when the system package manager has been used to install it as a part of the OS. It can be overridden with the flag --disable-version-check although I did consider hiding this under an obfuscated parameter name as the issue has come up several times now and because the configure process does not have the system library directories as its defaults.
Fixed a core dump that could happen without the X11 libraries and a window request that failed. With just the cli active this would always happen unless the -window option was given.
The Explorer and Voyager mod bus ramp was actually coded as a square wave. It's S&H was also very low. This also lead to more changes in the filter cutoff vs. keyboard tracking, it is now a bit less bipolar.
This version includes a manual page, a common one for bristol, brighton and startBristol.
Reincorporated code for channel pressure recognition. It had been dropped from an earlier release due to dumping core when applying the channel change to all the voices associated with that channel. Also fixed up some issues with the poly pressure handling, I think it was want to hit 100% CPU. This was ported back from 0.60.
Fixed a bug that had been reported a while ago against the MG1/Realistic of corruption of the GUI image and loss of sound. The GUI corruption was due to incorrect bounds checking on some bitmap manipulations which were corrected, a fix that is applicable to all the emulators. There were a couple of other specific issues with the MG1 and RhodesBass that had other consequences, also fixed. This was ported back from 0.60.
Added in another velocity map: no velocity sensitivity. That should have been there from the first implementation of maps but I think I got carried away with the ability to add hundreds of different curves that this got swept aside. It can now be requested with '-curve 0' to give something like a -3dB constant velocity level for all struck keys. Also back ported.
Added a play mode to the CLI. It can be activated with ':set play' and provides an alternative key path directly to the frontend to send midi note events. The code is only monophonic due to the way the raw interface works - there are no qwerty key 'off' events so they have to be interpreted. It does allow for quick testing of patches without having to resort to having a keyboard attached to the engine. Exit from 'play' mode is as with the others, either escape or ':' which at the moment cannot be remapped, something that might be an issue for internationalisation.
Andrew requested monophonic sustain pedal support. This does not really fit into a monophonic instrument: if all notes sustain then with note precedence the only note heard will be at the outer one. Using sustain with a polyphonic assignment algorithm over one voice works. For the HNP and LNP another algorithm was added that holds the other extreme note - the lowest with HNP or the highest with LNP - this is the note that will fall back when no other keys are held during sustaining notes. When a more extreme note is played it then becomes the 'extreme' note.
Julien reported some lost note events which was a bit disconcerting. Reproduced an issue where the events were not lost but their triggers could be: if a noteOn event arrived it would set the trigger flags. If a noteOff occured in the same sample period it would clear these flags which is incorrect for the monophonic note logic. (to be confirmed). This required a couple more patches for glide errata reported separately and largely related to inversion of the glide rate when the pitch wheel was moved causing Inf to be generated.
Colin Fletcher requested that something be done about the device names in jack and eventually ALSA midi. The fix is not straightforward as the engine does not know a priori what it has to emulate when it opens the MIDI interface. The current feature is that the engine recognise the -emulate [synth] and use the synth name here for all registrations. Added to this another alias called -register [name] which will set both audio and MIDI devices names to the given name and can be used to override other settings.
Extended the CLI midi command set to include 'set midi controller [id] [value]' to give access to the MIDI controller interface. May be of limited operational use but as a debuging tool it is very helpful - with alias supported it can be used to generate NRP, RP and the rest. This was implemented to debug Andrews issues with CC #64 (sustain) and used it to also debug Juliens issue by the same manipulation of the sustain flags.
Cleaned up the build processes so that they really do not need X11-dev to be installed. The previous releases did require the X11 headers due to some stray ifdef in the brighton header files, even if --disable-x11 was requested. Now, for the correct build process it is necessary to define --disable-ximage too although the ximage is really informational only, ie, it would not be compiled anyway without X11 being enabled. It was left in as it gives an error message regarding the missing X11-dev headers. May remove that in the future.
Cleaned the code up to remove the size warnings. They were almost uniquely on printf statements. The only remaining warnings should be for some unused variables which can be cleaned up later, and my explicit warnings for some potential issues that may need to be taken care of (again, at a later date).
This release is primarily a cleanup to the CLI, first released in 0.60.0. There were a few bugs that became evident, some incorrect cursor positioning and hence occasional garbled output. A couple of CLI features were also added which include extensions to the completion and the ability to add commands to the accelarated keycodes in ESC mode.
Finalised the CLI command completion code such that it can pretty much complete the whole tree with reasonable help information available.
Changed the CLI alias variable replacement such that variable $i is the param of that index from the alias command line, the previous syntax was a little bit unusual where a $ would just take the current variable index and % would make the variable increment.
set alias trans Osc1 Transpose=$2; Osc2 Transpose=$1 set alias trans: Osc1 Transpose=$2; Osc2 Transpose=$1 transpose 1.0 0.2
This will set Osc to 1-Octave transpose and Osc2 to 5-Octaves
Added the capability to set CLI key codes in ESC mode to call any :insert mode command. The interesting feature is to be able to configure aliases in the CLI and then give then single key accelerator codes:
1. set cli q quit
2. set alias panic set midi panic
2. set cli p panic
3. set alias clidata set cli
3. set cli ? clidata
The first will allow the use of 'q' to exit the emulator, trivial example but the point is that the CLI :insert mode command 'quit' is now available in ESC mode. The second one is potentially more useful, it links the ESC mode char p into a MIDI panic event which causes alls- notes-off. The third one adds a help function to escape mode: if you now type '?' it will list the CLI accelerator keys and their functions.
This cli/alias feature can also be used to provide seach accelerators for selected synth parameters:
set alias vol Volume
set cli v vol
Now when in ESC mode the single key 'v' will take the Cursor to the Volume control rather than having to /search for it or use the Cursor left/right motion. Several of these can be constructed to allow fast navigation through the parameters, one to jump to filter parameters, another to envelope, etc.
Added some options for changing parameters from the :insert mode in addition to the [Variable]=[value] that was implemented in 0.60.0. The new modes are '+' to increment by 1%, ++ to increment by the 'set accel' value, and +=[value] to increment in given steps. Decrement have similar actions too:
mini mem 0 chan 1 # Tune=0.5
00 Tune : 500
mini mem 0 chan 1 # Tune+
00 Tune : 510
mini mem 0 chan 1 # Tune++
00 Tune : 610
mini mem 0 chan 1 # Tune+=0.09
00 Tune : 700
mini mem 0 chan 1 # Tune-
00 Tune : 690
mini mem 0 chan 1 # Tune--
00 Tune : 590
mini mem 0 chan 1 # Tune-=0.09
00 Tune : 500
Some of the same syntax was added into the 'set brighton device [p] [o] [val]' such that where previously value was an exact value it can now take +=[f] and -=[f] to give relative changes to variables. This has some value for a few of the synths (poly800, bit series) where different panels may be required for data entry and parameter selection. The Poly800 has now been populated with text names for its panel parameters however to make it useful please contact the author to get hold of a suitable set of aliases for changing the data entry controls.
The ARP Odyssey emulator now has a 'Dual' option on one of the VCO. This will make one VCO assign itself HNP and the other LNP to get two note logic. This is not really dual voice as there is still only one set of filters and amps. It is not a feature of the original, it kind of emulates another emulator. To prevent unexpected results the code is not applied when in polyphonic operation as it relies on the monophonic note logic to decide which are the highest and lowest notes being held down. The feature is gimpy but more fun could probably be had by starting two independent Odyssey on the same engine and midi channel then applying HNP to one, LNP to the other.
Several changes were again added to the log file processes to remove a race condition on file open status. This only showed up on multicore configurations.
Patched up some compilation directive errors for FreeBSD distributions, also removing a stray ALSA header that should have been placed under a compilation ifdef for removing its support. This was primarily for debian support.
Bristol now has a command line interface available from '-cli'. If the code is built with --disable-x11 then it can be compiled without any X11 dependencies. When built with X11 then specifying '-cli' will open both the CLI and the GUI.
The CLI is based on a VI type interface and will respond to the arrow cursor keys: each parameter is available with Left/Right, and it's value can be changed with Up/Down for accelerated control.
When started it is in escape mode and will also respond to the following keys:
CLI: h left
CLI: l right
CLI: ^k incmin
CLI: k inc
CLI: K incmax
CLI: ^j decmin
CLI: j dec
CLI: J decmax
CLI: M memUp
CLI: m memDown
CLI: r read
CLI: w write
CLI: x toggle
CLI: / search
CLI: u up
CLI: d down
CLI: U fineup
CLI: D finedown
CLI: : insert
These mappings are saved in the synth profile and can be edited to change the them: '~/.bristol/memory/profiles/[synth]'. These mappings are additional which means the above values represent the default mappings that always exist. There is a keyword 'none' which will disable any key operations that you do not want. They call also be configured from the command line with 'set cli [k] [action]'.
Typing ':' goes the the VI style command line in insert mode and will accept the following commands:
GUI commands:
find: [free|load] synth memory search
read: load synth memory at the configured index
write: save current synth settings to memory
import: [path [mem]] read a file and save to memory
export: [path [mem]] save synth memory to external file
help: this screen
set: [line|debug]
midi: command set to engine (X)
debug: [on|off] CLI and engine debug settings (CLI only)
bristol: [cont/op/value|register] - operator commands
brighton: [panel|p/d/v] - GUI management commands
cli: [list|[key] action] navigation key management
alias: command aliases, %/$ signs will be remapped parameters
quit: exit application, terminate the emulator
The command line has a history buffer of the most recent 50 commands and any consequtive duplicates are supressed. When in ':insert' mode then the cursor motion keys will nagivate this history and give command line editing. The keys '^p' and '^n' are also history forward and history backwards respectively. There is a variable called 'set history [n]' which will limit the number of history lines that are ever displayed, either with the 'history' or the 'set history' command, this is for small displays. The command line can be edited with ^W (kill-word), ^U (Kill-Line), ^A (start-of-line) and ^E (end-of- line).
The shell command line history can be re-executed with :![n], :[n][TAB] will complete the line from the history buffer and leave it available for editing.
When in ':insert' mode the [TAB] key will list available commands and complete them where possible. Commands strings do not have to be complete to be executed, they only need to be unique. If a command cannot be found from the above list with the [TAB] key then the current line is used as a lookup into the synth parameter list with the following syntax:
: LFO[TAB]
: Osc 3 LFO
: Osc 3 LFO=0.5
The tabbed search here is for the given string anywhere in the synth parameter name, not at the start of the name. If the string is not unique then the matches are listed.
When in Escape mode it is possible to search the emulator parameters with the VI '/search' key and TAB is also available.
/lfo - find all parameters that contain the word lfo, complete if only one /^lfo - find all parameters that start with lfo, complete if only one
To toggle from ':insert' mode to ESC mode either type ^C (Control-C) or [ESC].
The CLI is available for the following emulators:
Mini
Explorer
Voyager
Odyssey
Axxe
B3
BME700
Realistic
Juno
Jupiter
PolySix
MonoPoly
OB-X
OB-Xa
Pro1
Prophet5
Prophet52
Prophet10
Sonic-6
Stratus
Trilogy
Sidney
Solina
Vox
VoxM2
RoadRunner
Rhodes
Rhodesbass
The arp2600, bm, bit1, bit99, bit100, dx and poly800 have not been converted and are for future study. The bit and Poly need a separate interface due to their input methods. The 2600 needs some method of representing the patch cable configuration. Now the CLI can still be requested for all of the emulators but only those listed above have (almost) descriptive text for each parameter.
There are a few CLI variables that can be set:
set history [n]
set savehistory on
set line [n] [CLI line width]
set accel [f] [Controller acceleration for Cursor motion]
The savehistory option will overwrite the emulator profile file with every 'quit' commands. The 'line' option defaults to 80 columns but is configurable for small Braille displays.
The CLI supports alias definitions:
set alias trans Osc1-Transpose=%
The commmand 'trans 0.5' will cause a transpose request to be made to the engine. Multiple commands can be aliased together using ';' between them. Multiple parameters can be set such as
set alias trans Osc1-Transpose=%; osc2-transpose=% trans 0.2 0.5
Here, the '%' is replaced with the alias options on the second line, each % will move the argument list forward. The command will send two different transpose events to the engine. Conversely:
set alias trans Osc1-Transpose=$; osc2-transpose=$
trans 0.2
Here the '$' does not move the variable forward so the same value is taken. This will cause the same transpose value to different oscillators. Any defined aliases will be added to the list of available commands and shown in the help menu, as will the list of [ESC] mode keyboard shortcuts. Aliases can be deleted with 'set noalias [name]'.
The CLI prompt can be configured. Typical values could be plain text for the emulator type but there are also variables available:
set prompt mini
set prompt %algo%:
or more usefully
set prompt "mini %memory%: "
will give
mini 99:
The tokens 'channel', 'algo' and 'panelid' are also recognised. It is also possible to give the name of any of the synth variables in the prompt too in which case they are evaluated:
set prompt "%algo% mem %memory% chan %channel% # "
mini mem 23 channel 2 #
set prompt "%algo% layer %PanelSwitch%: "
obxa layer 0:
Fixed an aged issue with the Explorer/Voyager that they produce no sound by default and even then they are very, very noisy. Not sure why it never got reported. The noise was originally due to injection to make the filters self oscillate but become noise after one of the many normalisation rounds.
After integration of the CV IO for the ARP it was time to change the diverse Keyboard Tracking inputs for the oscillators and filter. Up to now they had been inactive, the tracking was integral to the operators. Now they can be remapped from any of the other outputs which gave some facilities to get modwheel to frequency mapping and improve the filter keybaord tracking.
Change the MIDI device selection such that '-midi alsa' configures a SEQ interface rather than ALSA raw. This has caused a few error reports from people who have not got the expected result. Selecting a raw ALSA interface is now done with '-midi rawalsa' or '-midi alsaraw'.
After integration of the CV IO for the ARP it was time to change the diverse Keyboard Tracking inputs for the oscillators and filter. Up to now they had been inactive, the tracking was integral to the operators. Now they can be remapped from any of the other outputs which gave some facilities to get modwheel to frequency mapping and improve the filter keybaord tracking.
Several changes were again added to the log file processes to remove a race condition on file open status. This only showed up on multicore configurations. This was a 0.60 backported fix.
Backported the 0.60 fix for an aged issue with the Explorer/Voyager that they produce no sound by default and even then they are very, very noisy. Not sure why it never got reported. The noise was originally due to injection to make the filters self oscillate but become noise after one of the many normalisation rounds.
Patched up some compilation directive errors for FreeBSD distributions, also removing a stray ALSA header that should have been placed under a compilation ifdef for removing its support.
Since the release of 0.60 this stream is now in maintenance only.
Made the audio engine allow for registration of multiple ports into Jack. This gives the stereo in/out that always existed plus up to 16 additional ports which will be used according to the emulator requirements. In the short term this will only be the ARP 2600 which will register 4 extra IO for CV inputs. Future code will probably be the mixer that has been kind of waiting around for a while now. The option '-multi [n]' specifies the number of channels that will be opened to the jack daemon. The input and output signal levels can have a correction applied to them, -migc and -mogc each take a floating point which will adjust the gain on the signal, applying it to all channels. The default gains are 1.0.
The signals for this IO are normalised to +/-1.0f although the ARP shouldn't have issues with a wider signal range. The tests executed in house used two ARP both connected up to Jack, one driving CV into the other such that the Envelope from the first was applied to the filter of the second for example. The signal levels were within the range 0..1.0f from the envelope however AC signal will have a negative range too. That may be an issue for some other targets that do not want their CV to go negative so outputs 3 and 4 have a DC normalised signal with clipping below zero. Extensions to the CV configurations for the other emulators is for future study. The issue is not how to get the signals into or out of the emulators but a question of how the routing should be defined. For the ARP 2600 the signals can be patched as required.
A part of the IO functionality testing was done with session built by the Jack Session Manager so that two ARP 2600 could be loaded and signals passed between them. This lead to a number of fixes to parts of the naming structures, not really bugs, just things that would only really have come to light when running multiple instances of bristol with jack and jsm. The fixes included removing any auto-connection options if there were any jsm options present, renaming the windows so that the title includes the device name from the Jack interface to help identify windows in duplicated setups, more small adjustments to the parameters parsed to the JSM command line, some changes to the import and export routines used to load the JSM saved memories.
Recoded the GUI NRP support so that it can track NRP parameters. It requires the GUI be started with the -gnrp option. The code does not support ganging: tying multiple GUI devices to the same NRP however that may change presently. The existing CC control registration can be used for that purpose. The NRP support is naturally affected by all the usual limitations of NRP: they are assigned to MIDI channels not device targets (which SYSEX does do). This means if two devices are on the same MIDI channel it is up to the user to ensure that the NRP they use do not have conflicting actions. The NRP mappings are saved in the same profile as the emulator CC, CM and keyboard mappings. Per default the mapping table size has 128 entries which matches the CC counts and should be enough for any of the current emulators. A parameter -nrpcc can be used to change the table size.
The emulator settings will now be taken from the emulator header files rather than pure runtime switches. This imposes the use of the '-emulate' option which previously had this function. The choice was based on a discussion with Andrew Coughlan: if we are emulating then consider defaulting the closest match to the original, users can override options afterwards. This means a Hammond will start with the maximum voice count available, for example, but a Mini Moog will start with just one voice, highnote precedence. Changing the number of voices after definition of the emulator can make the Mini run as a polyphonic instrument if desired. The algorithm search is strict, ie, only the emulator names from 'startBristol -summary' are recognised. This is in contrast to the remaining parameters that do take some aliases (-arp is the same as -arp2600 for example).
Needed to suppress a few more options from the command line presented by bristol to the Jack session manager. These were port, rate and count which are all dynamic parameters anyway. Not supressing them lead to their duplication and some undesirable side effects of that.
Renormalised the trilogy output signal after some advice on their levels. Changed the configure flags to include options for separate default audio and MIDI drivers - Jack audio and ALSA MIDI was quite popular.
Fixes some repainting issues with the vertical sliding controls, they left a trace of noise on the way down.
The code now can actually save and reload sessions on request of the jack session manager API. 0.50.4 will be delisted.
Added a -import flag which will take a file and load it into the synth at init time. This may fail if the memory was for another emulator however it can ease file/memory interchange. The memory is not directly saved into the bristol cache but still needs to be saved to some (free) memory location in the emulator. The flag is actually an alias for -jsmfile which was required to do the same for the jack session manager.
Added in runtime options to support keyboard splits. Not sure why this took so long as it is a base requirement but as it is also possible from a master controller it never made it yet. The two options are -highkey and -lowkey and apply to single emulators. Error checking is minimal: values outside of the 7-bit MIDI note range are not checked, low ] high is not checked, ie, it is possible to build configurations that will not work. They will not break the emulator but no response will be heard.
Installed the svn jackd version with session management support to start work on coding the interface. It required a few changes, not just to the way that the jack interface had to work but also to the bristol message passing to allow for arbitrary strings to be exchanged rather then just control and operator data. Similar to the LADI code, the session requests come into the engine, it then has to take care of redistribution to the GUI where save operations are actually executed. The interface is perhaps not quite what JSM expects but that is purely a bristol internal working issue: Jack session requests come into the engine only, they are translated into bristol session requests and then distributed as required - to the GUI. Bristol session management includes the capabilities to both save and recall setup however the save and recall source can only be the bristol cached content, not from an arbitrary file. to overcome this, when a save request is sent from the engine to the GUI, it extracts the file path. Memory is saved to a private entry in the cache and then this private file is moved (exported) from the cache to the target location. The load request works similarly although Load requests do not come from JSM which does not have this feature. The engine command line parses the -jsmfile option and sends a session message to the GUI to import and then open it. There is the -jsmuuid option which is used by the session manager to direct the aplication to register with the given ID. There is one additional option which is -session which will disable all session management, including LADI.
Removed the incorrectly remaining 1/2 sample phase correction from the reworked Huovilainen filters for the non-resampling algorithms. These are called when the samplerate exceeds 80KHz as the point where the filter quality starts to degrade is far above the audible operating range of the filter. These have not been widely tested and an override could make sense.
The OB-X filter code needed some corrections for the performance changes, the code was there by the command line options were different matches. Also added some more frequency corrections for keyboard tracking of the filter.
Resolved an error in the jack ringbuffer management. The buffer would fill up and give 'ringbuffer exhausted' messages under specific circumstances: running as a headless server with last GUI detached whilst still receiving MIDI events. The fix was belts and braces, the buffer is reset in the audio thread if it is idling without emulators and the ringbuffer is also stopped with the last of the emulator to exit, restarted when new ones are created.
The GUI still had some legacy support for controller registration against things like RP, NRP, DE. This is supposed to be disabled if -nrp is given however the flag was not being honoured. This is a kind of workaround, a bigger fix would be to correctly register them however that leads to some issues with table sizes that will be left for future study. This whole area is a grizzly part of MIDI and should really be resolved with an alternative protocol.
The default build is now without the semaphores, they can be re-enabled on request at configure time. Also added a config option for the default drivers, the can default to alsa or take an option to use Jack as the default.
The mini emulator GUI would show key motion for midi channel and channel+1, this was caused by an incorrect key panel flag. The engine did not respond on both channels, just the frontend.
The Rhodes emulators stopped working due to some code that looks for the skins not functioning on their internal contructions.
Corrected (most) of the previous restrictions on band limited oscillators, namely that they were precomputed tables rather than dynamically limited. They did give different signals than the mathematically built infinite bandwidth waves (-blo 0) however they did not really prevent aliasing at the highest frequencies. The code now does cut in bandwidth limits depending on a few run time parameters:
-blo n - maximum number of harmonics to map, defaults to 31 -blofraction f - upper frequency limit as fraction of samplerate (0.8)
The code will attempt to stuff as many harmonics up to 'n' such that the fractional limit is not broken. Using values ] 1.0 will lead to pretty wild aliasing, lower values will thin out the sound but exhibit fewer artifacts at the higher frequencies. The default value is admittedly above nyquist, this is a tradeoff between distortions and waveform content. The code uses precomputed tables for all frequencies where freq*n does not exceeded f.
Reworked the Huovilainen filters to move the expensive maths into the feedback loop only. This reduces filter CPU requirements by about 90%, naturally at the cost of some quality. The net results of the changes do not degrade quality that excessively and due to the amount of reports of CPU exhaustion there are new filter options: -lwf are the chamberlains, -hwf are the original heavyweight Huovilainen and the default are the optimised ones. These are all run time parameters per emulator. There is still the overriding -glwf option to enforce all the emulators to use the lwf option globally. There is further filter option which is -wwf, these use about half the CPU of the -hwf filters, they use the same overall architecture however use a different but also more efficient non- linearity giving a 50% lower CPU footprint. The default filters may also be requested with the option -nwf, normalweight fiters.
Added some new corrections for the filter frequency tracking and especially keyboard tracking changed considerably. It can now be played in tune when at full resonance and can self oscillate at some settings.
The window now has a zoom accelerator under '[Shift] Enter'. This will toggle between native window size and full screen. The native size the the one at which it was created including applying any scaling parameters. The code will introduce some image antialiasing when it scales to full screen, this remains after returning to native size.
Used the same code to have window scaling accelerators: '[Shift] +' will make the window about 10% bigger, '[Shift] -' will make it similarly smaller.
With the zoom accelerators coded it was then possible to add in the -autozoom feature such that when the mouse enters the synth window it is maximized and when the mouse leaves it is minimised. Both windows remain active however they can reduce screen clutter. Width and scale will define the minimum and maximum window sizes.
Added an option -width to speficy the minimum starting window size. This works in conjuction with -autozoom and -scale so that between them they set the min and max resize limits. There is no corresponding 'height' option, the height of the window will follow the aspect ratio for the emulator. The option only works with autozoom, if that feature is not required then the -scale option should be used to set the actual window size.
Fixed the responses to FocusIn and FocusOut, they were silently dropped in the earlier code releases but were now needed for the autozoom capability. With these in place the opaque resize aspect ratio management was changed such that the resize results in the window getting morphed to fit the window and then aspect ratio being enforced after the resize has finished.
Glide and pitch bend did not work together: glide was cut as soon as pitch was altered. The fix was a one line change to the frequency assignments, previously target and current were set by the pitchwheel, the change was only to adjust the current.
There was an issue with logging to the user home directory, in most cases the log file would be opened truncated and never used as destination for output messages.
Added code for Jack Sample Accurate replay. This was a compromise for a useful if awkward feature to implement in bristol: it is only implemented in the bristol envelope code such that note_on events will be sample synchronised to the offsets reported by Jack MIDI events. All other events will be period synchronised as the Jack MIDI events are handled in the audio stream so are at least guaranteed to occur within the given period even running at double speed, etc. There is no intention to synchronise all the audio generation code and as this is working as an emulator that is not considered relevant - the original instruments all had free running oscillators and as such were did not have any kind of synchronisation to note changes. We are not going to attempt to have the exact same sound come out at the exact same point however the major state transition at note on will happen at the desired timepoint. Some of the emulators already had trigger synchronised LFO, these are now also sync to note offsets from Jack although the net effect will probably not be much due to these being low frequency hence low rate of change over the Jack frame offset reported in the note event.
Placed the use of sem_init under a complation flag, preferring sem_open instead, improves portability except that sem_open requires root privileges and PAM does not seem to support any security permissions for this. To use the sem_open call then './configure --enable-sem-open'.
This last change was probably superfluous as code was also added to remove the necessity for the semaphores in the first place. They cause issues with the RT threads and so incorporated a ringbuffer for note events from the MIDI thread to the audio thread. This adds a little extra processing but not excessive. This is a compilation time option, './configure --disable-semaphore', it will eventually become the default sequence or potentially put under run time options. The code was not completely trivial as the system needed to be able to distinguish between where events came from, and if jack then checks were needed to see if it was from a single registration or multiple registrations - the former actually ends up with two threads that can recieve messages, the MIDI thread fielding events from the GUI (including note events) and the audio thread itself - the sample ringbuffer code only supports singular TX/RX of the ring. Added some flags to the messages buffers to indicate source and make the choice of direct handling from audio thread messages or indirect handling via the ringbuffer under other circumstances.
Coded a second ringbuffer for event forwarding. To ensure low latency then it was not advisable that the RT threads did event forwarding to TCP however this is needed to get event tracking from engine to GUI. The second ring buffer is for the path from the audio and MIDI threads to the otherwise idle parent thread who then does the forwarding from the ring buffer.
Fixed an issue where inability to find the bristol profile files would result in total loss of output signal. The cause was failure to default the velocity and frequency tables when the configuration files were not found.
Removed some compilation flags that were specific to my system ready for distribution. Version 0.50.1 probably had compatibility issues due to these.
Implemented code for LADI level 1 compliance. This required a few changes, not just to accept SIGUSR1 to save state but also to have this possible in the engine and GUI. For the GUI this meant adding startBristol code to exec the GUI (or engine for that matter) rather than fork it, then have it trap the signal and save a state file. For this to work in the engine, seeing as it does not maintain state, it requires the signal be trapped and later a message sent to the GUI to save the state. This last one caused diverse issues that needed to be resolved to clean up some SYSEX handling. The code will accept options to field LADI only in the engine, only in the GUI, or the both. The default is to field in the engine. There is an option for the state file memory index. At the moment bristol does not keep an arbitrary state file: the call is a tack into the existing memory save routines so that memory, chord and sequence information is also saved, this code section only accepts an integer which is the memory index. If both -ladi memory location and a -load options are given then the LADI state memory will be loaded if it exists, otherwise the -load memory will be activated and then becomes the template for the LADI state file.
Reworked part of the Jack audio library on request of Nedko for LADI support so that the client interface is only opened once, in the audio thread. Without this he was seeing both audio and MIDI threads opening an interface and this did not map into his registration routines and PID mapping. Previous releases incorporated separate Jack registration requests in the audio and MIDI threads using different symbolic names. This worked however was a bit cumbersome for LADI support. The MIDI code can now reuse the audio registration from the audio thread, just adding an extra MIDI port to it. This became the default Jack behaviour with a -jdo option to enable Jack Dual Open if still needed later. The operation of the -jack flag also changed to configure Jack drivers for both audio and MIDI. Prior to this the -jack flag only configured the defaults for the audio interface and kept the ALSA (-midi seq) option for the MIDI interface. The previous functionality can still be requested with the flags '-jack -midi seq' or just '-audio jack'.
Integrated several keyboard accelerators, these were requested as a set of usability enhancements. The include ^S, save, ^L, reload, ^X, exchange two memories, ^K, print the list of shortcuts, ^H/^?, print help information which is the text writeup for the emulator. The actual readme is now built from this output. Parts of this code (specifically the save routines) were used for the LADI state save routines.
Cleaned up some issues with the MIDI interface flags, the readonly input to bristol needed to be configured as an output into ALSA, something that was being mixed up. This may be ported back to 0.40. Affected midithread.c and midiSeqDevMan.c only.
There was a bug in the memory file management code that cause a spurious directory 'memory' to be created in the users home directory if the emulator did not have a private memory path. It was also created in $HOME/.bristol as well, correctly, however the code did not return on the first creation.
The BRISTOL_LOG_CONSOLE environment variable was changed to be a true/false flag. The run time equivalent option -console should now be honoured under all circumstances which was not the case previously.
Cleaned up the closedown procedures for the situation that jack decides the process is 'zombified' (sic) when it is actually under CPU pressure. It used to leave the engine hanging around. The issue was not always obvious due to alternative TCP port selection but it was still wrong. The processes should now exit more gracefully although jack could be a bit more flexible here as the cause of the original hanging processes was libjack blocking disconnects from a process it had flagged as 'zombified' (sic).
Added key handlers for left and right arrow to give more controller tracking. The Up/Down keys already moved controllers, with shiftkey accelerators, the Left/Right now give more movement than up/down.
Integrated another fix for loss of keyRepeat functionality on exit: there was a window where repeat could be turned back off if the mouse re-entered the GUI window before engine negotiation had completed.
The following changes would have been in 0.50.0 had it been released:
This is actually a new software stream, the changes required to give correct distributed processing were sufficient to be kept separate from the 0.40 release even though there do not appear to be a great number of features.
This release 0.50, is going out first as unstable. Many fixes for bugs found whilst working on this code were put back into the 0.40.7 release which is now considered to be stable. Having said that, this release is the first to include some form of quality assurance. The process of SQA rotates around the distribution capabilities: a single engine is started in one window, and then a scripted procedure runs sequentially through every emulator, starting and stopping every 10 seconds (roughly). During the whole process note events are being sent to the engine and redistributed to the GUI. The SQA test is only considered to have passed if it runs without any form of misbehaviour. Typical cycles are somewhere over 2000 iterations. Failure includes any crash the GUI or engine, loss of audible signal, exhaustion of any of the library resources (handles, devices, etc), memory leakage or any recognisable memory corruption. This is neither the simplest nor the most complex test procedure. The first iteration just restarted a single emulator 1000 times, after that it became starting all of them, and then adding note generation. There are two other scenarios that will be in a final test sequence (at a later date) which will include multiple synth pairing and also program change events. The multiple synth pairings will be interesting, it should involve generating notes to every possible pair of layered synths (about 1200 cycles) including stopping and starting all of them each time on the same engine.
Added a -log option that causes output redirect to a file. Will first attempt to log to /var/log/[process].log if permitted, otherwise it will attempt ~/.bristol/log/[process].log with file truncation if logging is to the home directory. As a part of this operation the whole logging mode was altered such that a separate thread now takes control of stdin/stdout and stderr, timestamps the output messages before printing them. Finalised this with a -daemon option that detaches the controlling terminal and implies -log and -server. It is noted that -daemon holding open files in a users home directory is potentially not optimal so this mode would best be used as root with open files in /var/log. It is possible to override the use of the log thread by setting an environment variable BRISTOL_LOG_CONSOLE, this prevents the thread creation and continues to output to your TTY.
Had to rework the brighton thread management to allow for asynchronous MIDI event distribution. It may need more testing so will go out as unstable - there were quite large alterations required but it lead to better general thread management in the GUI: it no longer exits forcefully on window requests or interrypts, it flags the threads to exit gracefully. The changes all passed through the SQA hurdles discussed above.
Finalised the event redistribution code so that only the GUI or the engine need to be aconnected to a real midi interface (or sequencer interface). Events arriving on this interface are redistributed to tcp ports, ensuring that the GUI and engine always track the same events. There is an override of this functionality using -forward (discussed below) and forwarding is on by default.
Added an option -sysid to specify the MIDI SYSEX system ID for communication. This needs to be entered as a 32 bit hex value. It will override the default of 0x534C6162. This followed an interesting discussion on LAU regarding controller identifiers. The theoretical collisions that could happen should not have affected bristol since SYSEX is only used over the TCP connection from GUI to engine however the feature was cool enough to get implemented. For those who are interested, the default SYSID is 'SLab' in hex, the bristol MIDI interface extended from the SLab design.
An issue was reported with -voices 1 not giving any sound. This turned out to be true and the workaround (now the fix) is to use -hnp or -lnp with either the -mono flags or when voices=1. The code also now configures the note precedence automatically (although it can be overridden). [And the last note precedence code should also now be fixed too. nc.]
Fixed a spurrious compilation issue when Jack drivers are not found. Code was still attempting to make a library call that was not compiled into that library.
Reworked exit conditions so that the processes can return reasonable status back to the caller. That was not the case with the GUI and even if it was then the startBristol script needed to still consider these exit status.
Did a cleanup of the build process to remove some spurious warnings that should not have cause any issues with the code but did make the pedantic compiler churn out messages. The remaining output should primarily be for my own warning statements and a few unused variables/routines in the unfinished emulators.
Added an option -window to prevent posting of the GUI window. This is for SQA testing where each emulator is started in turn in a constant cycle of thousands of iterations, this spares a few cycles and allows other work to take place without the sometimes irritating window focus changes.
Inserted some exclusion code on the midi signalling not to dispatch callbacks for physical midi events, nor forward them, during some critical code sections, notably starting and stoping emulators. Also had to clean up some deallocation of the voices as there was a 'hole' in the code where a NULL baudio would prevent a voice from being freed up, ever.
Added some NRP commands to request remote event forwarding. This is used by the GUI to prevent the engine from duplicating events into two possible TCP pipes (dual manual keyboards), and by the GUI to selectively do the same for its interfaces. This resulted in changing the -forward flag to be a global engine setting, -localforward for the GUI to the engine (local in the sense that the GUI is per definition local, the engine may be remote), and then a -remoteforward option for engine to GUI settings. The latter two will only affect the single emulation which means some GUI can request copies and others decline them. This flag does not affect the existing -tracking option which simply governs whether the GUI keyboard will respond to MIDI note on/off messages. The forwarding of events covers more than just note on/off, most notably program change and modulation events are needed even if tracking is turned off.
Fixed an issue of incorrect shadow rendering from resizing withdrawn panels. When they are finally exposed the shadow contained the unsized instance.
Fixed two issues with the bassmaker operation, the first step after pressing start was not being sounded and LED status is not cleared on restart. The LED status of the last Stop point was not being cleared. Additionally the shadow rendering was faulty when panel selection changed with resized windows.
Cleaned up a compilation flag that could still cause it to fail if Jack was not included in the system or build process.
Reworked the audio library so that ALSA devices are just drained and prepared when a read or write error occurs. Previously they were closed and reopened which is a rather ugly overhead, plus there were obvious chances that it used to incur some memory leakage. The new code changes some of the data preload processes which now have to be pushed down into the library. The previous method can still be compiled by giving the --disable-drain option to the configure script for cases that give problems with the new method.
A failure in limits checking lead to a potential memory corruption issue in the continuous controller code and NRP handling. This could have been the cause of the original issues reported around NRP causing the engine to fail. Irrespective of this, though, there are too many possible collisions here as the NRP address space is common across a given MIDI channel making its use rather hazardous.
A couple of the emulators were exhibiting fairly random segmentation faults, indicative of memory corruption. Ran up valgrind to give some input into the causes and there were diverse issues that had to be taken care of. A part of the cause was the controller ID damages given above, there were diverse changes to the emulator code as there were a few errors in buffer pointer management that resulted in some emulations leaking quite voluminous amounts of memory each time they were invoked, and some small leaks on things like audio device restarts where buffered name space was not being reutilised. There was one case of a bristol audio structure being reused after it had been freed, this was introduced when the emulator destruction code was moved from the MIDI thread to the audio thread for other unrelated problems. Valgrind still reports some minor memory leaks however these are almost uniquely related to the ALSA library interfacing where structure ownership is not really that clear and are typically bits of memory that are not really lost, they are just not freed on exit, and since the audio drain feature was added then the audio device is not longer reopened when xruns occur. These remaining bits can be reviewed later as the loss is still small, it does not appear to increment over time and does not affect functionality yet. In total there were about 15 changes related to corruption rather than leakage although not all of them would have caused segmentation faults as they were not all write operations, some were read. These would have caused intermittent noise (ticks) and potentially wild modulation, especially of the DX FM operators.
Removed an issue with the libbristolmidi ALSA SEQ interface handling for the channel identifier, it was not correctly being merged across resulting in all channels converging onto zero.
There was a small timing window where GUI failure could result in dangling emulators (headless). Active sense will only start after a sense message is received, so if this first message does not get there then the emulator does not detect GUI failure. The engine now always starts ActiveSense, the GUI has to turn it off if not enabled.
There was a ludicrous amount of debug messaging added to track down all of the above problems and rather than remove it all for the release, they were buried under a -debug flag that defaults to 'off'. Values of '-debug 12' or greater are not advised as they will debug every period of samples.
The B11 library stops key repeat on window entry and reenables on LeaveNotify, this is use to make sure the QWERTY keyboard tracking functions as expected. With some window managers (reported against fvwm) the window Destroy function only sends a wmDelete request and no LeaveNotify as the window exits which would result in loss of key repeat. The command 'xset r on' fixes that however submitted a fix to turn repeat back on under these circumstances too, and added in some X11 event debuging at -debug 9 or higher.
Finally continued to resolve what I would call minor issues although that is subjective. Engine compilation did not include config.h, the Solina had very different gain on its harmonics, some of the bitmaps needs improved drop shadow rendering, the MIDI processing would terminate with the first emulator to exit, ie, the MIDI library was not reactivated, etc.
Removed potential segfault in midiNoteOn(). Put the speaker holes back into the ARP 2600, vertically as there was a shortage of realestate but having them puts the gravity back into the image.
Fixed an issue with the flagging defaults for the -alsa flag, it was taking ALSA rawmidi rather than the usual target of ALSA SEQ. This would fail the option completely. It could still be made to work with -alsa -midi seq, or by not using any specific driver options as the defaults were correct.
Updated diverse parts of the GPL compliance, bringing all the files up to version 3 of the license and including the disclaimer events where necessary.
Bristol version 0.50.0 was never distributed, the release was rolled up into 0.50.1.
Fixed an issue with the flagging defaults for the -alsa flag, it was taking ALSA rawmidi rather than the usual target of ALSA SEQ. This would fail the option completely. It could still be made to work with -alsa -midi seq, or by not using any specific driver options as the defaults were correct.
Jack MIDI opening sequences had a potential window to fail in which case the interface would not be activated.
There are no further fixes from 0.50.1 that will be ported back to 0.40.8 as they are now primarily related to new features in the 0.50 stream.
The 0.50 stream will also be created at the same time, as such this download may not be of interest to many people, ie, installing this may duplicate the effort later.
Fixed some compilation flags that could still cause the build to fail if Jack drivers were not included in an installation. This is a backport from 0.50 for compatibility purposes.
The sid.c code had as issue with debuging being executed without suitable checks on the handle. Could cause a crash which has now been fixed.
The XPM reader had a couple of issues with parameter naming collisions and its file naming for temporary files used for decompression could also fail if more than a single GUI intialised at the same time.
Had to implement some extra local memory sanity checks before calling the emulator operate() code. Depending on timing these may still be null as the emulator initialises.
The bristol BassMaker would give incorrect panel rendering when panel selections were made on a resized window. This could have affected other emulators as well.
The BassMaker would jump the first step of a sequence.
The BassMaker did not correctly clear the LED status of the stop position.
There was a hole in the voice management code that would prevent a voice from ever being freed up if it had a null baudio. This could happen if a voice is still active as an emulator is terminated.
Cleaned up a compilation flag that could still cause compilation to fail if Jack was not included in the system or build process.
A failure in limits checking lead to a potential memory corruption issue in the continuous controller code and NRP handling.
Removed an issue with the libbristolmidi ALSA SEQ interface handling for the channel identifier, it was not correctly being merged across resulting in all channels converging onto zero.
A couple of the emulators were exhibiting fairly random segmentation faults, indicative of memory corruption. Ran up valgrind to clear up a few memory leaks of small amounts of memory on things like audio device restarts and also to track down the corruption. It was partly the controller ID damages given above however the baudio structure was getting freed before its ultimate use which could intermittently lead to further damage. There are still some minor memory leaks however these are related mostly to ALSA library interfacing where structure ownership is not really that clear. This can be reviewed later as the loss is still small and does not affect functionality yet. In total there were about 15 changes, not all of them would have caused segmentation fauls as they were not all write operations, some were read. These would have caused intermittent noise (ticks) and potentially wild modulation, especially of the DX FM operators.
The B11 library stops key repeat on window entry and reenables on LeaveNotify, this is use to make sure the QWERTY keyboard tracking functions as expected. With some window managers (reported against fvwm) the window Destroy function only sends a wmDelete request and no LeaveNotify as the window exits which would result in loss of key repeat. The command 'xset r on' fixes that however submitted a fix to turn repeat back on under these circumstances too, and added in some X11 event debuging at -debug 9 or higher.
Removed potential segfault in midiNoteOn().
Resolved a name resolution issue that would cause the split/layer keyboard emulators to fail. The dual manual ones should have worked. The workaround was to default the hostname when null but the actual fix is to request the same destination host in all cases, something that will now only go into 0.50.
Included some memory packs for the trilogy and polysix emulators.
Added an option to startBristol, -gui, which will prevent the GUI from being started and invoke the engine with all the resolved bristol variables. This will allow for easier distribution of the applications with GUI and engine on different hosts, this was always a part of the design but was never really made that accessible. This can be used in conjunction with the -server flag which will leave the engine active even when the last emulator has disconnected. Running as a server lead to a few issues that were anticipated - constantly reconnecting GUIs would exhaust various engine tables that were not being cleaned up correctly as this had never been widely used. The file descriptor table was not being cleaned up, the handle and device tables similarly. There was an issue with SID selection of the EXIT requests not being correctly matched causing the wrong emulators to be disconnected, and with failure to correctly close input file descriptors on active sense failure requiring a small change to the device reading logic. It was finally possible to run a test script to reconnect several hundred times without failure, all emulators still audible and so moved to released code.
Several issues arose whilst building the application without ALSA drivers. This is supported, especially now that Jack MIDI is integrated, however since ALSA turned 1.0 the coding has been a bit lax, assuming that ALSA was available and integrated. The result was that building without ALSA might fail and that even when compiled there were issues with device open() requests. Resolved these and added the flags required to accept note events over the control link so that GUI keyboard events are still tracked even if the ALSA code is not integrated.
Was requested to remove all OSS dependencies from the build process as well. This was a little more work than ALSA since that was always intended to be a build option, OSS by contrast was always anticipated to be present. Had to clean up some of the socket toolkit header files and some erroneous definitions in the midi library that was being flagged by some more meticulous compilers.
Reworked the TCP addressing code to accept -host
There was some general code cleanup from the request to have bristol function on a system that had neither OSS nor ALSA installed. This was never anticipated and making it possible brought other issues to light.
The Trilogy had a very unequal mixing section, the organ gain was too quiet and the synth section too loud. These were evened out.
The Trilogy also had some rather unusual use of volume controls. To reduce CPU load the synth would not run the organ and string sections unless they had a non-zero gain. This achieved its affect however the result was that if the gain was zero at note_on then the voice would be muted (it actually just fell through the note logic and turned itself off). The fix was to always run the oscillator divider circuit envelopes under all circumstances to engage the note logic. The audio generation code is still skipped as it would be silent anyway when the gain is set to zero.
Added an autodetect for Jack where a small program will connect to the daemon and find out the sampling rate and period size. These are then given to bristol as parameters preventing unexpected mismatches later.
Fixed a watermark issue with the ALSA drivers, the high available threshold was too low which causes the library to report false overruns. Since these come back as a failed write operation the bristol audio library was restarting the audio devices. The diagnostics reported this as a broken pipe.
Resolved a typo in the ALSA drivers that damaged the periodsize and buffersize matching. Caused cyclic ticks in the output stream depending on hardware and ALSA driver versions.
Changed the activesense period to 3s and timeout to 15s. There are issues when running with RT scheduling where the GUI may get starved when there is a lot of audio activity causing the engine to give a false positive of a GUI failure. Since active sense is really there to make sure everything exits gracefully when there really is a failure then the timers need to be more flexible to cater for situations of high CPU load.
The Bristol B3 was exhibiting stuck notes, fairly arbitrarily however the issue was not reproducable on the development system. After a ludicrous number of debug builds to get different statistics the cause turned out to be the B3 postOpts which manipulated the voice flags and the resolution was to extend the semaphore coverage to include emulator postopts. Theoretically it would be possible to have just changed the B3 code however the engine that hosts the emulator should not be open to flag abuse in the voices and the voices are allowed to manipulate these flags. The actual cause was a race condition that only really exhibited itself on multicore/HT systems.
Thanks to Andrew Coughlan and Damon Chaplin for the report and debuging output to isolate this issue. As the problem was not reproducible on the development system (single core) then tracking down the cause required a number of different debug revisions and a great deal of patience and output logging from the people involved.
Releases 0.40.1 and 0.40.2 will be removed from the download site.
The bristol shutdown procedures were not really compliant with the Jack API definitions, they would not deactivate/unregister the active handles before exiting. This does not work very well with multiapp environments, it causes a subgraph timeout in the API. Changed the shutdodwn code sequences so that the last exit operation will clean up the emulators in the audio thread and let the MIDI thread then unregister jack and exit so that all parties are happy. Also added some diagnostic output in the event that ports cannnot be registered - this happens if Jack has a lot of application ports to deal with so the additional message makes sense. Also coded in jack_client_open() to replace the now deprecated jack_client_new() previously being used.
The Sidney emulator was about 1/2 a semitone out. After a lot of delving into the correctness of the code without much success (ie, it seemed correct) it turned out the C64 had different versions for NTSC and PAL, with different CPU clock speeds. The frequency tables were built from data taken from what had to be a confused manual since they did not match up for the clock speeds. When the due correction was applied (0.985 to 1.02 and finally 1.023MHz) it was finally pitched correctly.
The trigger events for the SID MOD Env were not really correct. Any accepted note_on event would cause them to trigger however in light of the way voice separation happens it really required that the Env was only triggered if it would have a affect on the given voice. There were a few cases, voice goes via the filter and env modulates the filter or if the Env modules the voice that is being activated. Added the logic. Voice-2 arpeggiating never triggers the envelope, only its own envelope.
Increased the minimum Sidney arpeggiation step, it was a ludicrous 0.3 ms, it is now a more reasonable 16ms. Maximum stays the same at roughly 250ms.
The Sidney emulator had an issue with some memories and its keymode settings, the indexes were not converged correctly resulting in the radio buttons not working normally (double selected radio buttons).
The OB-Xa 4-pole filter was incorrectly mixing its feedback loop, silencing the output signal. The 2-pole worked correctly.
The MIDI library was altered such that all messages now carry a sequence number, it's only u32 but its only use is debugging events and with the typical event rates this will last far longer than any session. Also finally fixed the event timestamps which were previously just null.
The midi and audio threads had different scheduling algorithms, one was FIFO the other RR. Since the two threads used semaphores for a critical code section for note events then there existed the possibility that the midi thread could cause large delays to the audio thread by taking the semaphore and then getting pre-empted. The fix was pragmatic, both thread have been made FIFO with different priorities. A full fix would also have been to adjust the thread priorities in the critical code however the current fix works except when there are multiple RT programs running and that should only be the case if they are all audio threads.
This is primarily maintenance for the last couple of releases including fixes to the SID emulator and some ARP improvements.
The SID synth has two more key assignment modes, both arpeggiating, visible in the GUI as Arpeg 1 and Arpeg 2. They will split the keyboard at MIDI note number 52 then assign two voices to one half and one voice to the other. Arpeg-1 has its function on the upper half, leaving the lower half as a duophonic synth allowing an arpeggiated chord as rythm for a bass sequence. Arpeg-2 will do the opposite, it will arpeggiate on the lower half allowing for arpeggiated rythm with a duophonic lead solo. These were seen as an improvement over the previous but the results need to be reported back.
Bristol will now check the system for the availability of the TCP control port, defaulting to 5028. The application behaviour has always been that an engine is attempted every time the application starts and correctly speaking if a user wants to work multitimbral then the second request should be given the -engine option. Without this flag it still worked but was sloppy, the second engine would fail to get its control port and so would exit, the second GUI would still connect to the first engine and start a second emulator. The current behaviour is that each time the application is started it will look for the port availability and if it is already taken then an upwards scan is done from the port number looking for a free socket. The port can be specified explicitly with the -port option if fixed ports are required however this is not a real requirement if the -audiodev is used with Jack, for example, to specify the registration identifier for the new engine. If no engine is requested the GUI will just attempt to connect to the specified port although arguably it should check to see if the port is open - this is kind of done implicitly when the GUI attempts to connect to the port.
Extended the BRISTOL_AUTOCONN by adding the following environment variables, BRISTOL_AUTO_LEFT, BRISTOL_AUTO_RIGHT and BRISTOL_AUTO_IN. If these are set they will be searched in the relevant ports list and then connected up. This only happens in conjunction with AUTOCONN. On my systems this was tested by setting them to system:playback_1, system:playback_2 and system:capture_1 resprectively, and by leaving them unset. Will see how much use they are, since they are environement variables then to be useful they would have to be scripted.
Submitted code to repaint transparency layer devices when windows sizes are changed. This affects shadow layers and transparencies (such as the ARP 2600 patch cables). Prior to this the objects painted on to this layer were lost which was confusing with the ARP at least. The effort to do this was considered preferential to the alternative which was to fix the window size once defined, not allow resizing and just have the -scale option at startup.
There were some compilation issues due to the SID test program not having the correct dependencies. Removed sidtest from the distribution, it is no longer required, it was only used to exercise the bristol softSID chip until it was integrated into an emulator, now done.
Known issues are that the Pro-10 would fail on some systems. This is a very specific issue as some systems have been reported not to suffer from the issue. A workaround is in place however a longer term resolution will have to be sought and a bug is open against the issue.
This release introduces an emulator for the Commodore C64 6581 SID audio chip. The implementation has a digital access method to program the chip registers and an 'analogue' access method to extract the audio signal. The oscillators and envelopes are implmented with integer functions as the original was partly digital, and then the filters are in floating point to emulate the analogue components. Specifics of the chip will be in the README file presently.
Two of these chips were then used for a bespoke synth, the -sid. This is not based on any original design but exercises the SID emulator. One of the two chips gives the audio output, the second gives signals for modulation: one LFO, one Env and one noise source which can be as an input to a S&H circuit. This design uses a few different voice allocation routines, monophonic, three voice polyphonic plus some combination algorithms to give access to the high frequency arpeggiation used often in the C64 where a single voice would do a fast scan of several frequencies to give chords and widen out the sounds. Each of the three voices can be programmed independently.
Added an environnment variable, BRISTOL_AUTOCONN, which when set will direct the jack library to autoconnect its ports to the first IO found. This will not be much use for general users however it will make testing easier.
Fixed an issue where the emulations would fail unless started on midi channel 1. The issue was caused by the Globals settings using a system connection id rather than the negotiated midi channel for setting up diverse defaults tables.
Coded a 16 step, 4 page sequencer along the lines of the SQ-10 to generate bass lines, hence the pun. The overall feature set it pretty sparse since this is not intended to be a general purpose sequencer. Each step has a note, transpose, volume for the MIDI note and a Control option to send fine tune, mod or another MIDI note on a second channel, plus the triggers can be skipped to join notes. The control features have not been widely tested and may need some fixes. Also the fine tune and glide when applied to the Control setting are a little dependent on the emulator, something that can be altered if the is a demand. A full description is in the README file.
Colin Fletcher optimised the houvilainen filter code, first factoring out kfc to reduce the number of multiplications being executed, then continueing with completely factoring out v2, and since it was generally used as a divider function this was a considerable CPU overhead. This delivered a massive improvement in the CPU utilisation, not only of the filter but the emulator generally due to the filter being so CPU intensive. The factoring out of v2 was ingenious since it was not immediately evident from the code that this would have been possible.
Testing this also corrected some anomalies in the OBXa filter selection code.
About 5 superfluous variables were stripped out of the filter code to reduce parameter popping and actually improve readability providing minor improvements in the filter efficiency.
Applied noise injection code generally to prevent filter denormals with low to zero input signal. This is quite efficient noise generation code so should not have a big affect on the optimisations and prevent 100% from denormals.
Another optimisation was applied to the filter for higher sample rates which removes the internal oversampling. At the higher rates this delivers around 40% reduced load and should not result in any loss in quality since it only applies when playing at 88kHz or greater where the filter response is still good to over 20kHz without resampling. Not all the emulators will actually use this code modification since it is only in B_FILTER2 which is sparsely deployed at the moment, something that will change as the different modifications are qualified. The result of all these optimisations is that this filter code should be about as fast at 96kHz as the previous one was at 48kHz, and as the filter code currently uses the vast majority of the emulator cycles then it is probably possible to run any of them now at the higher sample rates with little affect on net load.
Applied some reduced EQ to the Hammond B3 Bright signal levels. This kept the same rough profile however the overall signal level was far in excess of the existing normal gearbox. This was required due to some big reductions in the crosstalk levels of the normal gearbox leaving bright as too loud. The changes to the crosstalk was required since the net amount was superfluous and although it brings some nice overdriven sounds it was excessive for both of the configurations to have that amount.
Bristol no longer implements any Jack port auto-connect per default. After a trail of submits to LAU there were very good arguments as to why having default connections always being applied is a dangerous feature. The auto- connect remains but has to be requested with '-autoconn' although admittedly this does make it a fairly superfluous feature.
The VOX Continental M2/Super/300 was silent. The global parameters page had zero set for all parameters that needed corrections to the memory loading code to force the save settings into the active set.
The Polysix emulation would clip excessively with the modgroup was routed fully to the VCA. The signal gain from the LFO would result in an overdriven output. Reduced all the respective output stage signal levels, which are still quite strong anyway, and also some of the tremelo depth when mod routed. This will have a minor effect on some patches.
Incorporated Andrew Coughlan's manual page for the Polysix emulator.
When using Jack as the MIDI interface it was possible that the engine would attempt to link audio to midi channels and vice versa. The requests fail and do not break anything but the default connections don't work so its not quite as plug-n-play. Added code to check for the port name and also to parse the whole list of outputs for the first two that will link up, similarly all the inputs later.
Resolved a strange issue with the ARP 2600 with respect to discrepencies between the GUI indeces and the engine indices. The problem did not show up on all systems and the root cause was being masked by a programming error in the engine using an incorrect scaling for parameters passed to it. The problem was not evident on all systems probably due to differences in type casting - the parameter is delivered as a float however the index is taken from that cast into an int with a range considerably wider than '1.0'. The fix also required changes for removing cables.
The ARP 2600 filter and amplifier outputs had a signal level far stronger than most of the other components, this came from reworking the general filter input/output levels to ensure it was being driven correctly. Added in a bit of normalisation code to even the levels up.
The ARP 2600 filter would also suffer denormals if all the inputs were move to zero. Almost to be expected due to the DSP involved in the filter. The initial fix is to inject about -96dB of noise if all the inputs sum to less than that. This has the nice side effect of making the filter self oscillating at high resonance, that was not previously the case.
Reorganised the ARP 2600 volume options. There was a single 'Global Volume' that just drove the output stage. This was broken in several ways. Firstly it was in the memory which was a mistake to start with. Then it did not function as per the original where this single control was called 'initial volume' and was a kind of drone level for the VCA. To correct this the parameter was separated into 3 controls: Global Volume which is not in the memory, Program Volume that is in the memory, and Init Volume to allow the VCA to drone. It is noted that droning in polyphonic mode is messy however not having it with mono mode did affect capabilities.
The BME filter tracking keyboard selector did not function to expectation, the tracking was either off, or honoured along with the mod group which was wrong as these are exclusive options.
The X11 library will now only handle single configure events at a time. These are typically window resizing that come by the boatload due to mouse motion and if they are all handled at once we tend to get active sense failure and the engine (then the GUI) exiting.
Reduced the B3 drawbar crosstalk values for both gearboxes. The previous values worked but the signal to noise (crossbar leakage) was excessive. The bright gearbox settings are still quite rich in harmonics giving a reasonably overdriven signal but the normal gearbox is now not so noisy.
Distributed Andrew Coughlan Polysix patch 23 and added his sample to the website.
Build a BME-700 emulator. This is a rather rare synthesizer and is probably the best example of where an emulator makes sense, it would be pretty much impossible to get ones hands on one of these. The emulator is in its first build and the oscillator may well need some improvements. Future work will also improve response with higher samples rates by removing the internal oversampling code. To test this emulator try the options
startBristol -bme700 -mono -hnp -retrig
This will give a monphonic emulation with high note precedence and envelope retriggers for any note change.
Added a velocity flag to the monophonic key logic such that velocity may then optionally only be taken from the first note in a legato sequence rather than from every note in the sequence. Option is called -lvel.
Monophonic triggers were a little broken, the current fixed code will either always send a retrigger when the voice moves to a different note with the -retrig flag or will only send a trigger for the first of a legato sequence.
The Korg MonoPoly emulator would crash. The new synchronisation code did not check for a null sync buffer and some of the emulators would pass a null buffer if sync was not configured. Related to this was reassignment of the -mono switch to be 'monophonic', equivalent to '-voices 1' rather than as an acronym for -monopoly. The switch was reassigned to make the monophonic note logic more sensible.
Prophet bank select was broken, code was adding incorrect offset into the selection algorithm.
The note assignment logic now correctly implements/emulates a monophonic keyboard with low note precedence (-lnp), high note precedence (-hnp) and last note precedence. Previously there had only been code for last note - it was the polyphonic algorithm with just one voice however this gives a completely different playing style. The option only works when the voicecount is set as a single voice. Additionally there is a -retrig option to trigger in both directions when playing legato style. The code is interesting since it was firstly a little awkward to do and also since the final solution was to have the monophonic synth start playing its single voice and then just keep the voice rolling forever. It stacks up the CPU constantly but gives some useful features with the ARP 2600 and Moog Sonic-6 (and others) for droning or just tweaking them without playing and is arguably a closer emulation of a real mono synth. The voice, once assigned to a mono synth, should not be preempted by other emulations, ie, it becomes dedicated.
A present release may make hnp and retrig the defaults for mono synths, and similarly invoke a single voicecount if precedence is requested. For not you will probably want to use -hnp/-lnp and -retrig to get useful results.
This release might include some monophonic note logic debugging output, it will probably stay there until 0.30.7/8 depending on testing or bug reports.
Finalised the frequency/step tables and moved on to implementing glide for the non-resampling oscillators. The oscillator already supported this but the frequency tables were needed to correctly fill the buffers. Poly-800 parameter #67 now controls glide for that emulator, previously not operable.
Reworked all the filter code keeping only the Chamberlain and Houvilainen. Then reworked the Houvilainen for 2pole/4pole (non-resampling versions later). Put in an Oberheim modification that remixes the different pole outputs back into the main signal path. The mix is configurable however the GUI has no control that drives it: a default value is taken. The filter is a bit richer for the mix. This mod was at least considered by Oberheim for the Matrix-12 but then never implemented as the discrete circuits with quality components was a bit cost prohibitive. Here is is relatively simple work and since the filter is already computationally expensive this operation does not add much to the CPU requirements. Different codes sections use different mixing loops. The main one puts the tapped signals into the output stage only, others will mix it back into the filter feedback loop which give yet another quality however it also suppresses resonance due to the resulting phase complexities.
[As a note, the application still implements 3 other filter types, the rather weak rooneys and a butterworth. The ARP also has its own lag filter for voltage processing.]
The Arpeggiator/Sequencer code would leave voices hanging after the operation terminated. This was kind of known, start/stop just set the emulator flags to allow the code to be called. The resolution was to actively deschedule all voices associated with the arpeg/seq function.
The LFO changes to 0.30.4 to increase the maximum speed did adversely affect a number of memories and, rather than back them out, this version introduces parameterisation to set the upper and lower limits of the LFO with defaults that were as per 0.30.3 and below. The new emulators that want to use LFO up to ranges that verge on FM functionality will request wider ranges. The default values should go from 0.1Hz to 20Hz in line with the previous releases.
Build options was damaging the LD_LIBRARY_PATH where jack libs were included before the newly built library path. The result was that rebuilding the code would probably find the installed libraries rather than the new ones.
The Hammond was loading excessive memories on switching the opts panels in and out, added some flags to prevent this.
Worked on a slew of new (short) recordings for the website, fattening out a few of the exising memories. This lead to some issues with sound quality in some emulators. These would have been fixed earlier if I actually used bristol however most of the work is in development rather than operation. Most of the sounds were configured in a matter of minutes, more work would have improved the depth but they give an idea of each of the emulations that were recorded. These short recordings highlighted a few anomalies in the code, and along with some other issues that have been gathering over a few releases the rest of this release is a set of fixes.
There were apparant envelope clicks on the axxe emulation, primarily on note off events. This was due to a direct 'gain' level being mixed in with the ADSR. Corrected it with a grooming envelope on the key however the ARP lag processor would also have done this trick quite well.
Applied the same change into the Odyssey as it uses a similar gain section on the amplifier. The 2600 is not affected by this since it uses patching to get around the amplifier rather than fixed signal gains.
PWM on the Arp DCO was failing, affecting all the Arp emulations. The cause turned out to be a mixup with the min and max limits for the pulse width, something that was introduced when the corrected sync code was introduced in release 0.30.3.
Fixed a key mapping issue where the '\' character mapped to key id zero rather than two semitones above the ']' key where it should really have been mapped. The changes are in the text profiles so could have been done locally in the event of complaints. Apart from that the mappings only cover a US QWERTY keyboard anyway.
Some of the keyboard graphics were damaged during the redesign for the Poly-800 giving incorrect redraws for the 3 octave emulations.
The Poly800 GUI had a typo that stuffed a non-zero termination signal level into its DCO-2 envelope causing clicking if Double was selected. A separate issue occured with the filter envelope that was not retriggering correctly, resolution was to configure this one envelope for re-zero and will have to see if it needs to be done for all of them.
The duplicated raw audio output code does silence suppression however this was only intended to be for leading silence. It was implemented as a cheap way to get sample for the website but it was doing permanant silence suppression which damages the audio in most cases.
Added a -mbi option and structure entry to the GUI for memory supersets that came as a side effect of reworking just some minor details of the polysix options (the -load worked however the GUI did not reflect the true memory). This was rolled into general use however some emulations already use local methods to manage this. This will only conflict if the global method is requested and it also affects Chord and Sequence memory as that does not access this parameter (yet). This is a single digit parameter adding that number of K to all memory access. Memory #1 with '-mbi 2' accesses memory #2001 on disk. The option is additionally set from the -load setting, if a memory above 999 is requested then mbi is stuffed with the most significant part and the memory index is the remainder.
Revised the LFO to have a range from 0.1 to 100Hz for a pending BME emulation. The rate control is a power function so it should not affect any existing memories excessively.
Broken in this release and until further notice is the -glwf option, it causes note vocalisation errors. The -lwf option should be used with each emulator to activate the lightweight filters individually rather than globally.
Built a Korg Poly 800 which was more work than anticipated. The GUI needed to have a 'membrane' support for the parameters which generally takes quite a lot of effort, perhaps less since the bit-1 already had much of the code. Another envelope was built as it is a 6 stage design however that is reusable for the pending Yamaha CS-80 that has been under construction for a while. The Poly800 emulates all the dual/single oscillator with an envelope each, plus the single filter design was integrated into the emulator PostOps() routine as well as the Poly code so that there can be a single high key tracking filter as per the original design or you can have one per voice as a bristol mod. The full set of mods and general construction are described in the README file, the only remark here is that the single filter code is a bit jumpy when the notes change by large amounts, something that could be improved.
Added modifications to the bitone oscillator such that it integrates a separate sync output. This is at the base frequency with a resampled squarewave only, allowing this complex oscillator to still drive sync into other oscillators. Tested with the Crumar and Jupiter emulators.
The same modifications were rolled into the explorer DCO since it also generates some combined waveforms that could distort the sync algorithms.
Reworked the bitone oscillator to generate what I am for now calling waveform tendencies. The output signal will tend towards a target value under a kind of discharge circuit at a controlled rate. The result is a non-resampling oscillator, more or less, the wave is generated based on algorithms rather than wavetables. The results are at least harmonically unique within bristol, at the moment only integrated into the Poly-800 emulator. They sound reasonable when heavily filtered, a little raspy otherwise. The nice thing is that they are still quite efficient (will change as the waveform heuristics develop), costing only a few percent for the generation of 8 oscillators each with 8 strands of tendencies, all slightly detuned from each other to give a quite rich result. The oscillators can generate square, ramp, saw, tri and sine with support for PW/PWM of the square, sync in/out on ramp/saw/square/tri. The triwave sync is correct softsync, the rest are hardsync with waveform inversion as required. The efficiency lends itself to oversampling which in turn could make the Houvilainen oversampling filter more efficient and give a net improvement in the sound at very little cost however that is FFS and will not be done until after the P800 is released.
Integrated a new noise generator which has better performance and also a cleaner overall signal. The previous code has some excessive LF components.
The Preacher gearbox was a sixth out of tune, 8 semitones. Pretty strange it was not reported and I admit I had not noticed it. The default gearbox seems to take the wheel indeces from an incorrect starting point. This also brought to light another issue with the same code. Since the user profiles were enforced recently then the tonewheel mapping file had to be in ~/.bristol and was not shadowed back to /usr/local/share. This would have caused some issues with the tones rather than the frequencies.
Built another set of keyboard layouts, this time for the Poly-800 but may get rolled into other emulators later.
Check Button devices would still dispatch an event if the mouse is released outside the bounds on the device. This should not be the case so checks were added for limits before callbacks are made.
There was a small window when killing the app during initialisation could leave the engine waiting for the audio thread forever. It should all exit so added some timeouts to the status polling routines in the midi thread.
Key bindings were not being read for library testing. Not an operational issue however it needed a fix for testing purposes at least.
The pitchwheel and its associated Registered Parameter were not working in unison, had to reorganise their internal variables and the frequency calc code when changes are made.
Reworked a few more shade layers and put the new key designs into a couple of the other emulators.
Configure will now prompt the user to install libasound2-dev as well as libx11-dev if either are misssing.
Added the String section to the Stratus emulator to produce the bigger brother of the pair, the Trilogy. As with the Organ section of the original, the string section is not a major feature and was nothing to write home about. The code adds some extra harmonics to fatten out the sound plus some panning and spacialisation to enhance the section. The features can be disabled to give a more authentic emulation of the original.
Changed the graphics on the Stratus synth waveform selector, it was not a true mix but a three way switch. The graphic is emulated however the control is still continuous here.
The default memory for the stratus had a bit too much key tracking configured and it shut off the filter completely for the lower octaves.
Fixed some minor issues with the Stratus options panel.
Took excess debugging out of Trilogy/Stratus code.
Touched up the shadow layer for both these emulations.
The Polysix mg controller side affected the oscillator frequency. The wheelmod should just affect the overall depth of the routed LFO, it was doing separate mod for vibrato which was undesirable. Also changed the default switch for the polysix from -poly to -polysix and -poly6. For now it will respond to both. This was an unfortunate choice originally and will presently be dropped to just the newer options.
Similarly changed the Korg Mono/Poly switch to be -monopoly from -mono, also an unfortunate historical choice.
The polysix waveform selection was actually wrong regarding Square and Pulse wave - square is a fixed wave that only uses a PW control to select its width, the pulse wave uses the same setting but adds PWM via another LFO, these were out of sync. Thanks to Andrew Coughlan for the insight whilst working on a Jens Johannsons lead sound, apparantly originally done with a Polysix.
Andrew's patch also showed some issues with parameter memorisation for the operating modes: Poly and Mono mode could be lost on reloading a memory.
Added in some diverse changes for the PWM code, historically the width has been an integer value in the range of the size of the wavetables and at least for some of the oscillators this was still being managed as an integer. The pulse width is now floating point and the resampling is independent of the main waveform being generated. The difference is only subtle however it did show up a couple of bugs in the prophet DCO code regarding offsets and brought to light the noise on oscillator sync with certain waveforms (notably square) that really also needed to be addressed.
Fixed what was a long running issue with oscillator sync, namely that it worked but very badly if the synchronised wave as a square. It was always ok with ramp for example. The cause related to how sync was being done, it was just resetting the resampled wave table index to zero, correctly speaking it needed phase inversion as well, now implemented. The synchronising wave is also resampled before it goes to the synchronised wave to remove the flutter or shimmer associated with synchronising to whole samples, it basically synchronises to subsamples. Phonically this is a big change and it may be rolled back into 0.20 since 0.30 is still relatively new.
Alterations for this fix are still needed for bit1osc, it generates complex waves by default and they are not very good for sync signals. For now users will have to simplify the synchronising wave, a future release will have the oscillator generate a sync output signal at its base frequency.
This release has an implementation of the Crumar Status 'duosynth', it used an organ divider circuit to produce some slightly lacking organ sounds with four harmonics of pure square waves plus a six voice poly synth. The organ circuit has been beefed up considerably - it can produce the same square waves however adds in alternatives (square and sine through ramp, smoothly) and some added stereo spacialisation. The interesting synth glide circuits are emulated, this is not typically polyphonic portamiento since this beast used organ divider circuits for the oscillators so independent glide on each voice was not possible. In contrast the circuits could be over and underclocked as a note starts then glide back to the target note, here emulated polyphonically. All the legato features should work - the LFO in mono mode will only strike open the LFO envelope grooming circuit once, the glide features will work in a similar legato fashion and the oscillator output can be kind of random with legato.
Improved the oscillator sync algorithm however certain waveforms are still noisier than they should be. This was only in the Crumar Stratus oscillator but the alterations can be moved over to the other synchronising oscillators when finished.
Altered the filter keyboard tracking for the Houvilainen to try and get better tuning. It can be made playable however that still takes quite a lot of tweaking so a few more changes are likely over the coming releases.
Put the reverb back into the Sonic-6, will see what people make of it as it only works well with fat sounds (B3) or at low delays.
Reworked some of the button control logic since they were not being set when withdrawn. The logic for withdrawn should really be just 'don't draw' and this was corrected.
Reworked some more off the graphics for better drop shadow on the keyboards.
Voyager Electric Blue released in conjunction with some fixes to the existing explorer algorithm as detailed below.
Moog Sonic 6 released, the suitcase synth. There are diverse differences with the original, implemented largely because they were denoted at the weakest points of this fair synth:
Added a mod wheel that can drive GenX/Y.
PWM is implemented on oscillator B
Installed an ADSR rather than AR, selectable.
No alternative scalings - use scala file support
The emulation is not duophonic. Primarily poly with separated glide. It does have the same 'diaphonic' capabilities although they were pretty dubious in the original. Details are in the README file. There are sadly few memories configured in this first release, none in fact.
The Sonic-6 implementation was the first new algorithm in a while and it has shown a few issues with gain levels regarding the filter output. This lead to a few more alterations to the normalisation of diverse emulators but it should fix the issue of filter resonance overdriving the outputs and causing clipping. It should now be a gentler easy sine wave, probably still quite strong but it should not cause clipping with most emulations.
The Explorer code needed to be reviewed for a changed filter set and filter characteristics, the original stacked LPF was wrong. The code needs to have either serialised HPF/LPF or parallelised (and then stereo) LPF. This was worked into the code and an interface put together to emulate the 'Blue Ice' version of the synth.
Reworked the Release parameter of the Voyager/Explorer. The existing code either configured the value or just '10' depending on the release switch. The correct coding is to either send the value or a small part of the value based on the switch setting, closer to the original.
OSS drivers were broken due to a failure to convert some of the device flags correctly. The result was no subfragmentation and no subfragment buffer. Only tested with ALSA compatibility mode.
Reworking the drop shadow step by step on the piano keyboard. Gives a better effect.
Minor changes to the options for the different Prophet emulators.
The 0.20 stream is now maintenance only.
OSS drivers were broken due to a failure to convert some of the device flags correctly. The result was no subfragmentation and no subfragment buffer. Only tested with ALSA in compatibility mode.
Altered the permissions on the memory files and pixmap files since some users would have had problem accessing them. That would have made memories unreadable and given some incorrectly painted images. There were no complaints however it probably would not have been totally obvious what the problems even were.
Changed the lightweight filter option to be emulation local rather than global but kept the global option now as -glwf. This allows for mono lead synths to have a thick filter and the rest to have the lower cost Chamberlains.
Added an option to prevent the user interface from requesting window sizes to the window manager by default used to ensure that the aspect ratio was some kind of reasonable match to the original however this causes problems with some tiled window managers causing endless redraws. The -ar or -aspect option will just make the library paint into whatever available space is given to the GUI - the results may look gruesome if either the height or width is excessive.
Added code to the XPM reader to look for *.xpm.gz and then to gunzip them first to /tmp. This reduces the installation nearly by a factor of 5 and is minimal overhead an this is only in the startup routines. If the bitmaps are manually decompressed then the overhead is avoided as the xpm is then found first. It probably needs to be noted that the benefit of having the gz images will only be seen if the previous installation is first removed. If not then the .xpm and the .gz will remain.
Alterations to the diverse blueprints, most notably the MS-20 however since that is not yet functional it was a bit superfluous.
The following synths still need some filter work as they both have two stacked filters, a high pass into a low pass:
-explorer - moog voyager.
-cs80 - Yamaha (unfinished)
The CS-80 is not yet complete, in fact far from it, however it will need the same feature and the Explorer is the best place to start the work so that the results can be tested in a working emulator.
Added some checks for the number of options being given to both bristol and brighton, and if they are too few to then advise using startBristol and exiting. The goal was to advise on what should be done if somebody just attempts to use the binaries without the wrapping script, and since the script uses a host of default switches it doesn't cause these messages to print.
Resolved a long running and annoying problem with closing the GUI windows - when the close is requested from the Window Manager [x] button (typically in the top right) the application does not exit as fast as it should. Had to set some WM attributes and do some more checking for Client Messages. Also had to clean up the code to terminate dual manual synths, something that became evident once the shutdown did not require a ^C in the controlling terminal.
Improved the state logic between the threads such that if the audio thread cannot get hold of the target audio device then all threads will exit. This was giving problems since previously the parent thread would not exit even though the audio thread had failed. Recovery was then a manual process. We do not need to apply similar logic to the midi thread since if that does not open we will never proceed to the audio thread anyway.
Known issues:
Starting two mini on the same engine will seg fault when one is closed. The buffer pointers are cleared incorrectly. It will probably be a trivial fix: not to free the buffers on exit. The floating buffers are little overhead and will get reused if another emulater is started.
Integrated the Houvilainen filters into the full set of emulators. Some still have filter options that may use the Chamberlain however that may change by converting the Houvilainen to a 12dB/octave option and using the Chamberlain for lightweight filters, BP and HP:
Added an option for lightweight filters where the CPU intensive Houvilainen are replaced by the colder but a lot less expensive Chamberlain, -lwf.
Bandwidth limited oscillators are now the default with up to 31 harmonics. The previous geometric waveforms are still available with '-blo 0' however their roughness was highlighted by the reworked filters. A few emulators do not use the global wave tables and will not inherit the BLO option but they should all use other distorts to reduce the edginess of infinite bandwidth waveforms.
Normalised the signal levels on most emulations. This was lead by the Pro One developments since it gave a reasonable signal and was used to baseline the other emulations.
Converted glide into a power function to give better control at low values. Converted attack into a power function to give better control at low values.
Pitch wheel depth was half a semitone out of configured value. The default of 2 was being misinterpreted.
Rhodes was damaged due to a conflict in operator selection since the code for the arpeggiator was finalised. The Chorus controls would select ARPEG code and that would eventually cause issues.
Noise was inadvertently mixed in the Mini algorithm, passing through the reused filter buffer. Fixed this and other anomalies with mod routing.
Mono emulators appeared to drop new notes when played quickly. The actual cause was a loose coding of the note_off logic where an off event would match any voice on the channel even though note_on had reassigned the voice a new key. This only happened if the note_on voice was still on the newlist waiting to be promoted onto the playlist.
Normalised gains and integrated Houvilainen filters to other emulations.
Built a Sequential Circuits Pro-1 as a polyphonic emulator. There are some differences in operation compared to the original. Details are in the README file.
The Prophet-10 needed some work primarily on layer management. The original had a few features for in-memory volume/balance/tuning, a sequencer and the memory bank functionality was organised differently. These were all worked into the emulation. The sequencer is a lot simpler than the original but can function with a sequence per memory and adds the ability to arpeggiate and to chord notes. Also reorganised the 5/10 voice management since it was a little incorrect previously and realigned the midi channel utilisation.
Reworked the hammond drawbar device to support fully painted images and used that to render a new ModWheel dev with a rolling tooth design for the Pro-1. It could be improved with a local highlight rather than using the global shadow layer, something for future study.
Reworked the rotary pot to support limited motion from -60 to +60 degrees with stepped movement for the Pro-1 octave selector. This could be used in the other emulations however the general use was for waveform selectors and the problem here is realestate to show the waveforms - without the full rotary action the diagrams will not fit the available space. Mono, Poly now also use it for octave controls and waveform selections requiring a bit of redesign of their blueprints. Also put into the LFO waveform selection of the Jupiter. The work also included redrawing code the complete rotary to support irregular designs. This was required for the tip of the ProOne pot that sticks out.
Added a parser for Scala .scl files to build a microTonalMap for the synth. It is possible to retune the synth based on the thousands of Scala maps. The option is '-scl
Added another two taps to the Hammond chorus organised such that there is a unique tap sweep per VC rate. Each sweep is offset against each other and for chorus then either 1, 2 or all 3 taps are remixed with the original signal at a configurable mix level. Changed the scan betweeen taps to be a more gentle sloped fade rather than a linear transition. The scan delay and LC are user configurable parameters and the tap scan time also to alter the overall speed of rotation. Altered the addition and subtraction of taps to fatten out and smoothen the sound. Reduced the spread of the taps since they are only interesting at low delays. Filled out the lower scan ranges so that they at least always scan between taps. It no longer sounds like it has a broken cap.
Added a build option to configure: --disable-jack-midi. There were complaints of interoperability with the older jack MIDI libraries. I do not want to support the previous releases and this is the lowest common denominator to allow the application to at least build with the 0.103.0 API. The complaint is "too few arguments to function 'jack_midi_get_event_count()'" or similar.
The audiodev name is now used for jack registrations. It is defaulted to the program name when jack drivers are requested and then overwritten if given as an explicit option. This allows visibility of each invocation of bristol into jackd.
MS-20 was packing too many panels. Cleaned it up and added in the new ModWheel however it still does not work - should move it along a little bit.
Minor bitmap manipulations for some resizing misfits using the -scale option.
Added some more output text for cases where jackd could not be started however a review should be made of the the audio and midi thread status selections to make the operation smoother.
Fixed some issues with detuning (sensitivity) where if not initialised it led to unexpected results.
Put in limits testing in bristolMidiSendMsg(), something I did not want to do preferring a well behaved GUI however it has lead to issues during development that would have been nicer to have avoided and its at very little cost.
The per user private memory cache should now be enforced, this should allow people to keep their private memories over reinstalls and upgrades. Prior to this the cache was only used if the directory structure was in place however that is rather unwieldly for most users for the current number of emulations.
Added in bandwidth limited master waveform tables for the base waves: square, tri, ramp/saw, and a sine wave just to be complete. These were then pushed into the diverse oscillator code for most of the emulations. The results are a lot smoother if a sufficient number of harmonics is used, '-blo 21' or more, however smaller values also give interesting results. The Mini is generally improved by not having the 'infinate bandwidth' geometrically correct waveforms, the sound is softer and warmer with relatively few harmonics. The next work on the sound generation will need to go into the filter.
GUI now sends active sensing updates to the engine. This means the engine can detect UI failure and exit rather than hang around and damage the next attempt to connect (with the same MIDI channels, etc). There is one timer to control the rate at which the GUI sends updates and another time for when the engine will decide to stop the emulation. The GUI can detect failure of the engine the same way - the write operation of the active sense will fail if the TCP connection terminates at the remote side. The code could be improved, it should actually scan the list of synths however it is pretty trivial to extend it later. The reason is that the heartbeats are per emulation rather than at the top level however the generation of the heartbeats occurs in the GUI in a separate MIDI thread and hence is at the top level.
GUI now has a -scale option to presize the window. Default remains 1.0 and at large scaling will automatically enable aliastype 'pre'. This can be overridden with -aliastype none. There were too many remarks about the diminutive size of the GUI and whilst the author likes the size this option improves usability.
Started work on a Yamaha CS-80.
Backing of Realistic MG-1 was offset since the prealiasing code, needed to be corrected (actually had a third layer still packed but not used).
Explorer glide was not working due to changes to the operator indeces in the emulation not being mirrored in the GUI.
Change the max device ccont from 128 to 512, already under the limit with some of the synths probably leading to a few of the crashes I have been seeing.
Applied build patches from Alexis Bailler's gentoo packaging.
With antialias and mobile devices (sliders) we need to repaint extra pixels during move operation or the 'blur' starts to stain the image. Fixed with some minor alterations to the Undraw code extending its range by a pixel on the X and Y axis.
XImage creation failure now suggests using the '-pixmap' option but its almost a case of better never than late. I could consider changing the flag and then attempting the load again however this may already be a non-issue (was a broken 64 bit X Server issue).
Introduced a lever operator that changes its size at it moves up and down its scale. This was for a part of the CS80 code.
Added a configurable timer into the GUI host code midi scanning period.
Horizontal scales needed redrawing fixes.
Integrated a 4+4 split/layerable Jupiter algorithm with arpeggiator, sequencer and chording function assignable per layer. Full write up in the README file and web page.
Changed the Jupiter design such that the VCO have selector buttons rather than single switches for register and waveform. This is not per the original but it allows me to make them non-exclusive and that will fatten up the sound. This functionality is more like the JP-6 and MKS-80.
This code includes a Jack MIDI client interface. It is started with the normal options plus '-midi jack' so that it can be used in parallel to an ALSA link and TCP link to the GUI. The result is that the engine uses two midi threads, one doing bulk operations for GUI settings which might be quite slow, a higher priority thread doing Jack MIDI, and the highest priority thread doing the audio processing. Semaphores are implemented to prevent the threads interfering with eachother. Feedback on operability would be very much appreciated.
Reorganised qwerty mappings for correct MIDI note selection and transpositions, brightonController.c confdev may need to be buried in window structures. Worked especially for a couple of keyboards that were not C-C.
Implemented a prealiasing code which uses whole pixel blending. It is only applied to the silkscreens and gives far better imaging when the synths are resized towards fullscreen - they are still reasonably readable. Reducing the window size and normal size are a bit blurred. Configured with '-aliastype pre'.
Resizing windows failed on some emulations with the output signal going to zero: If some of the GUI settings did not match the engine before the resize then the sound will change. Supressed all updates when resizing since no changes should happen to the engine.
The implementation of antialiasing and window size changes also threw light on some coredumps at very small window sizes caused by anomalies in the drop shadow rendering going negative due to the reduced offsets.
Reevaluated the arpeggiator step rates to go from 1Hz to 20Hz.
Arpeggiator did not correctly terminate the note list when more keys than available voices were pressed. Resulted in hanging arpeggiation and holes in the arpeggiation sequence.
Resolved an issue with the -emulate and its interpretation of gain levels, they were a bit low.
Touched up the rhodes bass graphics. Minimal, but was bored.
This is maintenance code and you may want prefer to defer using it until 0.20.5 is released. There are numerous fixed integrated here, the midi note event management has changed considerably and there may be some anomalies in the implementation of the multiple dual linked listed that govern key assignments. The code was heavily debugged to ensure the it was correct but the symptoms of such anomalies are potentially quite ugly if you run bristol with real time scheduling, the tight loops may lock your system so you may want to run it with '-priority 0' for no realtime prioritisation. There may be lots of thread and arpeggiator diagnostics that will also disappear in the next release but you can also use the -quiet flag to prevent them. Also, since the MIDI and audio thread now use exclusion then a very busy audio thread may result in what appear to be dropped notes: The note_on gets delayed due to system overhead but eventually generates a 'newlist' entry for the engine to process. If the note_off arrives before the audio thread has taken the newlist entry then it is removed before it even gets started. These actions are visible with the current debugging. A master keyboard should not be badly affected by this however the GUI can end up clumping together multiple events if the CPU load is high and that causes/exacerbates the situation.
Reworked the OBXa for some stronger sounds with a couple of routing corrections. Overall sounds is a lot richer. This lead to changing the glide to be a log controller rather than lin, another improvement that will affect existing memories though. Will become a general capability. Fixed some flagging issues with OBX modpanel and layer assignments. Adjusted the mod mix and corrected the mod LFO rate controller. Should build some more OBXa memories as well. Rework OBXa dual/split options for single midi channel. Reworked some of the OBXa graphics including a shading layer ready for distribution.
Arpeggiating sequencing now records under suitable control - only emulation with an interface is in the OBXa, Bit-100 but others may follow. The settings can be saved in a memory to be recalled and recoded to the engine. There is serious fun to be had now with the OBX-a emulator and sequencing/arpeggiation. Details are in the readme file but basically start arpeggiating and then crank down the filter envelope decay and play with the filter cutoff and mod level.
Implemented the chording code as a small extension to the 0.20.3 sequencer. When enabled it will assign a voice to every note in the note list, transposing them as it goes.
Worked the arpeggiator changes back into the Juno-60 code. Made the OBX-a Seq be once per voice rather than once per synth.
The LFO rate parameter was converted to a log controller rather than a linear one, it improves response and allows for some very slow cycles. It will affect some memory settings. Glide also changed to logarithmic control. Arpeggiator rate similarly.
The memory library was changed such that it honoured the active parameter when calling memory locations, something that may well have changed the characters of the other synths. Went through the code to clean up the obvious ones but should run through them all again.
It is now possible to have the GUI search for an emulation rather than passing the emulation explicitly. This is with the -emulate option. The feature also sets default voice counts, detunes, gains, etc, from the emulation such that the options such as "-voices 10 -emulate jupiter8" and "-voices 10 -jupiter8" are different - the first call would default the voice count to 8 from the emulator, the second would have the emulator inherit the explicit option of 10 voices and since we have not actually requested an emulation then the remaining options will not be taken from the application defaults.
Installed some new key bitmaps, not sure about the results as they look a little surreal, almost chrome rather than ebony.
Corrected some tv_usec comparison failures that were damaging the double click timers and defaulting to 1000ms only.
The bristol ASLA seq interface would misinterpret note on/velocity zero. Caused some strange results with certain keyboards, and only with some bristol features (arpeggio) since it is correctly reinterpretted later in the chain. Also, the OB-Xa midi tracking would double strike with the GUI linked up to the master controller, the GUI was not honouring exclusion and retransmitted midi events already seen by the engine.
Reorganise the transposition system for some emulations, they now correctly request the transpose to the engine rather than just scale their own keyboard. Both methods work and are arguably both correct however unless the engine is informed of a tranposition then it will not be audible from a master keyboard driving the engine. This alteration needs to be reviewed as it separates the midi key id from the qwerty keyboard id's and that is awkward for anybody who wants to build their own mapping files. The mapping is now not qwerty key to midi key but qwerty key to GUI button index - for most emulations this is the key 0 to 60 on a 61 notes keyboard. Also separated the mapping table per window rather than globally - when the GUI goes medusa this will be required.
Resolved another window timing issue with note on events at high load. This required using a couple of semaphores and was a big change to the note event management. It ironed out a few other unlikely situation such as running two monophonic synths simultaneously would give unexpected results - their note selections could get intermixed. The use of semaphores for mutual exclusion of the midi and audio threads was very worth the effort as it will probably be used to distribute load over different cores with multiple audio threads presently.
Resolved a lingering seg fault in the DX operator with the bounds checking that was added and cleaned up the L1/L2 gain levels to be lin rather than log, the log did not work well for these controls.
Fixed a heap corruption in GUI has needed addressing for rather a long time instead of providing a long term workaround. Resolved a couple of related issues with detection of pure blue from negative offsets that also caused segfaults that have been annoying me for a while. The workaround that had been in place for several releases was pretty ugly.
Cleaned up some parts of the code that were giving in my opinion spurious error messaging during compilation.
Finalised the bit-100, some extra options, ability to save -1 and -99 memories.
Restructured the rate algorithm for the glide, there were historical issues with its mathematical correctness. Additionally added a bristol NRP to allow the maximum glide time to be configurable. The default value was up to 30s however that often felt excessive and makes configuring short values awkward, changed this to be 10s with the ability to request up to 30. The GUI now responds to the -glide option per emulation and the Bit emulations have memory parameterisation.
Added pink and filtered noise options to the BIT synths, defaults to white. The option 'white' could be dropped and turned into Pink with progressive filter.
Added a flag required to enable NRP parameter changes in the engine. Default is now not to pass them for interpretation since it is a rather arbitrary interface and has been probably responsible for noisy interaction with some master keyboards.
Diverse fixes to the BIT emulations: Layer volumes were damaged going into split/layer settings. Layer transpose, split and midi channel tracking needed some corrections for the different options. Specifically there are flags for supporting different settings per layer (per default they remain in sync) and these were not being honoured. The same fixes were required to ensure that non-aligned settings would be correctly returned after dual loading splits and layers. Layer parking was damaged (required writeThru cache enabled to work). Added parameters for the 32' subharmonic and square wave generation. Memory loading was damages for high ordered parameters. More issues with the high order parameters.
Restructured the arpeggiator to centralise the code ready to put it into the Jupiter emulator. Code inserted to allow the arpeggiator to hook into the midi note event dispatching to prepare it for reprogramming.
Tested now with jack 0.99, 0.103 and 0.109 successfully.
Updated copyright dates to reflect 2008.
Built a Crumar Bit-99/Bit-1 and an m2, all variations on the bit-99. Added several bristol features via data entry parameters. The capabilities and the differences to the original are documented on the website and in the README files with the distribution. This is an early release of the code and the memories are a little spartan. That is a bit of a shame considering the flexibility of the emulation, having over 100 parameters per layer. Fully implements the splits and doubles, including the quirky left right voice assignments in stereo mode. The next upload will have more sounds and probably another interface to access more parameters directly. Due to the flexibility of the interface design it was possible to integrate more options than a fixed configuration would allow. The parameter list may also extend in future releases.
Edited brightonButton for highlighted buttons, required repainting the control and could be rolled back into the brightonButton rather than being separate.
Patched in a dummy audio driver such that no audio operations are executed. This may seem rather daft for a synth however working under virtualized systems often gives dodgy and unreliable audio interfacing so having a dummy lets me still work on the GUI to engine interfacing rather than limiting the work to the GUI. This is accessible with '-audio dummy'.
Fixed an issue with port redirection and multilayer synths failing.
NRP were being incorrectly interpreted which meant that emulation gains and detunes were not being registered.
Velocity of note off events was being mapped into voice velocity as well as note-off velocity. This is incorrect as it causes jumps in envelope gains, etc. The correct behaviour is for device that want to use note off velocity should search for it in the correct place. The bristol ADSR does not adjust release rates by velocity however it could by using this explicitly.
The envelope uses a power gain curve that is precalculated for lookups on attack/decay rates. It was a little damaged at the zero point and could cause 'inf' failures on some of the work being done, highly inefficient. Only seen in very rare cases and normally just during development since the code avoided using the limits. Corrected.
Implemented OMNI mode however only the Crumar allow access to it at the moment.
Transpose has been reimplemented such that the MIDI library is correctly handling the transpose. Previously it would lose track of some notes - press key 36, transpose 12 notes and it would not be able to match the note off to the original note on. The fix was resonable trivial and now the rest of the synths should also start using this method, something that will happen later.
Velocity control has been added into the GUI so that it can be requested at start time per emulation. I should try and document all the 500 or so different mappings - the only references at the moment are the source code.
Compilation without jack or ALSA would fail dismally, the make routines would not pull out all the ALSA definitions. There may remain some issues since the OSS Midi interface does not seem to work, only rawmidi. If you really want OSS support with MIDI and all then mail the author. The changes altered the list of drivers, removing the Sun D-BRI and ALSA revisions prior to 0.9, both of which are now antiquated anyway.
The remaining audio interfaces are:
OSS (pretty much any release as the interface is stable)
ALSA 0.9 or greater (SND_LIB_MAJOR=1)
Jack 0.99 or greater
There does not seem to be any jack version information available from the include files at compile time unfortunately, something that may be an issue with the currently new 0.109 release that appears to have altered the library interface specification.
This was not released until jackd with a 64 sample period size was tested (reported to fail) and we also the jack interface has been folded into the bristol audio library even though it does not like being there. The alterations to the audio library to remove the diverse unused interfaces and changes to the compilation flags seems to improve stability (and may be ported back to 0.10 at some point).
Cleaned up some of the debuging.
Release 0.20.1 is based on 0.10.13 with the above changes. Apologies to the people who downloaded 0.10.13 yesterday, this release followed rather shortly afterwards.
Resolved a failure in the Vox M2 percussive harmonics, they were not decaying, an issue with the operator selection in the GUI.
Cosmetic alterations to some of the vox switch images, prophet panels, etc.
Added some bounds checking to the jack interface to prevent a couple of errors that occur when failing to open the interface.
XImage issues with 64bit sytems - the XCreateImage() fails due to pad and line length errata. Workaround is to use -pixmap (avoids the XImage acceleration code) and the fix needs testing but does not break 32 bit systems.
This will be the last development of 0.10 stream, no more features only bug fixes. Further developments will go into the pending 0.20 stream.
RT priority is now defaulted to 75, a value of zero for the runtime parameter does not reschedule the thread at all, it is left up to the linked libraries.
Vox Continental would segfault due to pedalboard buffer pointers.
Implemented a workaround for damaged segment sizes with some jack attachments but this is not really a fix, it just avoids the problem by exiting a little more gracefully when there is a mismatch. It may become the fix though, the problem only happens when I cannot find the period size a priori.
Reworked parts of the MIDI library code such that it checks for data availability rather than going into a wait state. This is a lot cleaner as it has a timeout for failure situations. Timer defaults to a few seconds only, that could be improved.
Rhodes voices were broken due to reorganisation of the ADSR default gain to be linear rather than logarithmic, firstly the notes barely decayed but also the harmonic content did not vary from the FM algorithm.
Integrated a chorus into the Rhodes GUI and a wrapper in the engine FM startup code.
Built some default "temperature sensitivity" into the dual manual synths, it should improve the sound, at least when layered.
Reorganised the voice initialisation code to have better GM2 conformancy.
Double click timout made into a GUI configurable option.
Added a 7 segment LED digit block to the library for the Jupiter display.
Implemented the arpeggiator for the Juno. It was a requirments for the Jupiter however it is easier to test with a synth that already works.
Reworked the threeway switch so that it was sequential, required some changes to the emulations using them.
The frequency table for the VOX was 'too correct', rewrote the vox microtonal map to have a table based on integer division from 1MHz to introduce a few cents of detune as per original.
Some of the microtonal maps were incorrect, loss of key 108. Had no effect as none of the emulations used the microtonal map however it needed to be fixed.
Put a bass pedalboard into the Hammond B3 emulation. It requires extra graphics space which is not an issue, and for both emulations the pedalboards kind of come for free - they use the same MIDI channel as the lower manual but call a different oscillator depending on MIDI note. The B3 memory structures had to change to support the lower manual drawbars. If you had your own memories in ~/.bristol/memory/hammondB3/* then you should really delete them.
Made the audio thread priority a runtime option.
Reworked the prophet pot bitmaps for higher contrast and closer alignment.
Distributing the GM2 MIDI controller mappings as default, not all emulations support it though - does not work well for the dual manual synths nor the organs.
* EQ: control the gains of each wave according to a profile.
* EQ: drawbar profiles.
* Distorts: control the harmonics of the wave, also with a profile.
* Rework chorus (other file) - need to take another filtered delay algo.
* New reverb (other file). Reworked reverb for 2600/Aks too.
* Always pass through Leslie - have it stop rather than bypass.
* Added in random phase difference between teeth.
* Crosstalk: between teeth of same compartment.
* Crosstalk: between drawbars.
* Crosstalk: between box output filters.
* Crosstalk: Different setting of bright/normal.
* Drawbar leakage.
* Tapering of drawbar busses.
* Check foldback, seem to be too much - damaged for damping. (tapering)
* Drawbars to full 9 stage swing.
* Drawbars graphics numbered.
* Added permanant drop shadow to hammond and pro-10.
* Percussive bypass of chorus plus add in HPF.
* Fixed ALSA Seq registered device name.
* Fix parameter damage on lower manual. There were two, resolved.
* Put diverse defaults tables into a header file.
* Fix lost note issues. Timing issue between threads.
* Fix memory linkage failure on exit. Effects mem allocation error.
Most of the hammond extensions take two setting for 'normal' and 'bright', one of the front panel buttons. This gives access to two potentially widely different tonewheel configurations directly. On the GUI side there were some fixes and additional shadowing to some of the renderings.
This release still includes development versions of an MS-20 (graphics only) and a Synthi Aks.
This release still includes development versions of an MS-20 (graphics only) and a Synthi Aks.
This release still includes development versions of an MS-20 (graphics only) and a Synthi Aks.
This release still includes development versions of an MS-20 (graphics only) and a Synthi Aks.
This release includes development versions of an MS-20 (graphics only) and a Synthi Aks.
01/01/2007
There is no news other than the arrival of this page.