From the ALSA wiki
WildMidi 0.2.0 is now released. Go to http://wildmidi.sourceforge.net for more details.
WildMidi is written by Chris Ison and uses the same config file format and GUS patch sets as TiMidity but as Chris explains... WildMidi isn't based on TiMidity. It was written from scratch. The only parts of TiMidity it has anything to do with are the config files and patch sets. It does a lot of things very differently to TiMidity and, as far as I know, there is no common code. Although some discussions I had with people involved with TiMidity had common ideas or topics, the way we approached most problems was different. It's currently only available via CVS from...
cvs -d:pserver:[email protected]:/project/cvs login (hit enter) cvs -d:pserver:[email protected]:/project/cvs co wildmidi cd wildmidi; ./bootstrap; ./configure; make; make install
Some notes from Chris's emails...
NOTE: There is a known bug in the reverb engine where it produces a lot of white noise at a sample rate of 11025Hz. Sample rates of 22050Hz and higher are fine. This is due to the unique nature of the reverb engine within WildMidi and I will have to look at alternatives to fix it. The reverb engine is only roughly tuned so it may not sound as you expect, but it works.
WildMidi is in 2 parts:
wildmidi (the command-line executable)...
libWildMidi is the core of WildMidi; it does all the hard work and is
designed to be used in other programs. For example, QuakeForge has a
plugin that uses libWildMidi (without reverb) so that
MIDI files can be played through the Quakeforge sound
wildmidi is a demo program that is just a command-line player.
-c on the command line to point it to your
wildmidi -c ~/eawpats/timidity.cfg my_midi.mid
The library only uses Gravis
.pat files, as used by
TiMidity, and also uses the
basic MIDI setup. It does not support the XG/GS extentions that
TiMidity does. However, it has extensions that
were added to "fix via config" some bugs in some pats that were
constructed incorrectly (with editors that didn't support the 3 release
envelopes that Gravis
.pat files contain). If you have samples that
sound wrong, let me know the name, size and where they may be found and
I will endeavor to give you the config settings to play them properly.
If you use the command line program
wildmidi, these are the keys to
use while a MIDI file is playing...
r switch reverb off/on n next midi q exit - decrease volume + increase volume v switch between curved and linear volumes (see below)
(this is because some MIDI files created on computers were mixed with
editors that used linear volumes instead of the curved volumes most MIDI
hardware uses. In some cases you won't notice a difference) You can also
turn off reverb from the command line, just use the
-r option on the
Unfortunately I do not have a mailing list. I do however hang out with
the QuakeForge crowd in #quakeforge on irc.oftc.net if you care to drop
by for some discussion. Many of them help test
wildmidi and one person
has even created an
xmms plugin for it (dunno if its valid for the
current version). As stated earlier the wildmidi program only
demonstrates a small fraction of what the library can do.
The library can...
- Be used in multithreading applications
- Process multiple MIDI files (which it does in QuakeForge) (in QuakeForge you can have 2 different MIDI files playing at the same time using the lib)
- No note cutting ... [(TiMidity](?title=(TiMidity&action=edit&redlink=1 "(TiMidity (page does not exist)") can cut notes when CPU load if heavy, WildMidi does not and with some MIDI files can give better sound quality than TiMidity due to this fact)
- Super-Fast Sampling ... WildMidi actually loads and processes the sound samples as they are required. This is able to be done due to the speed at which WildMidi does the re-sampling and mixing. Also, when run without reverb, there will only be a small amount of CPU usage.
WildMidi does not support all MIDI messages at this stage due to the complex or slow code required. ie: tremolo and vibrato are not supported at this time but may be supported at a later date. If there is something you would like to see in the program or library, feel free to suggest or even discuss what you have in mind. WildMidi is a work-in-progress and started out as a project to understand MIDI and software synths. But as you can see, it's grown by demand since then and I'm more than happy to add functionality to it.
gus pat format is actually quite flexible; it's just mis-understood by many who implement it.
I've done many many months of research into different sound formats,
sf2 (and even some
mod formats). Basically, guspats
are raw audio files in 8 or 16 bit format. Their sample rate can go as
high as 65535Hz (16bit rate entry) but you'll often find that the sample
rates are between 22050Hz and 44100Hz for size economy. Each file can
have 1 or many samples, each sample being tuned to a specific note.
What often occurs is someone has a pat to which they want to add more samples, and will add more but not at the same quality (or a slightly different sound) so you end up with bad ranges in some files. (one of the saxophone pats in eawpats has a good example of this). Also, what can happen is the software used to edit the pats may not understand the 6 (six) envelopes gus pats have and thus set envelopes badly. (This is an issue and I've had discussion with TiMidity and pat authors previously trying to tell them that gus pats have 3 release envelopes and not one, but they said their editors only showed one).
WildMidi uses the 3 envelopes as intended, going by documentation managed to scrape off the internet (and with a little bit of playing with original GUS pats). Unfortunatly pats made for TiMidity may sound wrong in their released version so WildMidi has extensions that you can add to the timidity.cfg file to fix this problem.
Example: in my gravis.cfg
0 acpiano amp=125 remove=sustain env_time3=4000 env_level3=0
This is REALLY bad in TiMidity. Its piano sound doesn't decay naturally. It drops so far then holds until it is released, which is very annoying and wrong. What those settings do is to remove the sustain bit (which tells wildmidi and TiMidity to hold the note at a certain amplitude level) and also ensures the 1st release envelope lasts 4 seconds with the resulting sound level being at 0 after that time. This may need tuning of course but that's the example I use (to make sure it's still working). Basically WildMidi allows you to edit the envelopes within the pat using the .cfg file itself. Each sample within the pat has a 32bit length value (in bytes). WildMidi however has a 32bit max filesize limit (compiler and operating system restriction which is limited even more in wildmidi by the ammount of RAM your system has available).
-v option to display debug info, if wanted, would be nice
but the progress meter and able to accept keystroke commands while
playing should not be the default (IMHO).
lol, debugging this thing can be evil to the ears, a single printf in the main resample/mix loop kills the audio completely, but I've used it at times to check stuff like the reverb.
wildmidi -h does not show the
-r option... and again, the
default should really be off.
Sorry about that, 'cause it's a work in progress, some options have
changed and I haven't put them in to the
-h yet, ie: there is no
surround option now, the reverb engine naturally puts in a surround-like
What about the same MIDI file playing back to two or more instance of wildmidi each with a different set of soundfiles ?
Ok, I'll explain this more. The wildmidi library is setup to handle as many MIDI files as your system would allow you to throw at it (memory restrictions basically). It uses the same configuration file for all the MIDI files you throw at it, and each MIDI file is handled individually without affecting the others. This is achieved with the use of unique file handles the library passes back to the program each time you load a file into the lib. If you want to process the same MIDI file more than once in one go, you have the library load the MIDI file as many times as you need, recording each file descriptor the library returns. WildMidi then loads each one into memory and treats them all separately so there is no cross-over of information and no MIDI file can affect another MIDI file.
Note: you can pass the library the filename of a MIDI file, or the
memory address of a complete buffered MIDI file in memory (has to be in
MIDI file format). If you do the later, you don't have to worry about
freeing that area, basically it's a "set-and-forget" 'cause the lib will
do the rest. ie:
malloc() some memory, load a MIDI file into it, and
pass the address to the library -- no need to free it cause the library
then assumes ownership. As for different soundfonts, you would need to
have different instances of the library running. I haven't tried it but
I don't see why you couldn't do it. You could use different banks.
WildMidi will accept up to bank 127 (standard MIDI banks).
I have the std and eawpats sets and am currently trying to run TiMidity twice in ALSA-mode sequencer so I can use sounds from either set at the same time (like the same instrument at the same time from both patch sets = fuller sound).
Just playing the same MIDI file twice with a modified program through the library will achieve this but you may want to reduce the volume of the mixed output it feeds back to you before mixing together because unchecked clipping is hell on the ears.
Is wildmidi anywhere close to being able output more than stereo ?
It would not be that difficult to add additional output channels to wildmidi, although the legalities of it could come into play in an opensource project. ie: Dolby's Digital Surround is very possible in WildMidi, yet patents and licencing issues are a problem so it's on in there.
[Comment from Greg Lee] I don't think you need to generate Dolby Digital. At the low end of computer surround setups (that's my end), the soundcard is hooked up to six (or eight) powered speakers with 3 analog stereo jacks, and you just send a six channel interleaved stream of samples to the kernel driver just as you send a two channel stream. There's no Dolby encoding involved. At the high end, I gather, people want to use home theater type equipment which does require a DD signal, but they need a sound card or motherboard with a DD encoder (presumably duly licensed). In any case, your midi player doesn't have to do DD encoding. In fact, I've modified two versions of timidity to do 4 or 6 channel surround sound. The Alsa driver and library on Linux know how to pass on 4/6 channels to my soundcard, an Audigy2 zs. It works.
The reverb engine in wildmidi is a combination of 2 different types of all-pass filters, with some low-pass filters as well. I can't remember the types of filters but basically it uses all-pass filters in a matrix setup whose outputs are arranged in a cross-over array switching the echoes from left to right and vice versa. As I said it still needs tuning but for an initial experiment it worked much better than expected. Basically if I can find out ideal decay times for the different reverb types (room, hall, ...) that I've seen on keyboards (I don't actually own any MIDI keyboards -- all my stuff is done on computer) I could fine tune it better.
I also want to learn MIDI and software synth architecture so I am very grateful to be able to stand on your shoulders with a better view than what would otherwise be available :)
Beware the pitfalls of the broken MIDI file ... there is so much sanity checking done in wildmidi when it's given a MIDI file to play it's not funny. WildMidi will refuse to play broken MIDI files (even if it's just a track missing). There is no automatic correction and I don't plan any because auto corrections can be very wrong when it comes to MIDI files, and that missing bit of MIDI information could have been crucial.
There is still one problem I'm hunting down a solution for although I really haven't looked for one yet as it's not critical. A couple of MIDI files I have for some reason have really large delta times near the end of them (delta times are MIDI timing units) so when the library calculates the total playing time, it's WAY off. But it plays the MIDI file fine, it just doesn't stop when it should because of the extra long delta time at the end :/ (I really hate hex editing of MIDI files but I've had to do it at times to make sure that what wildmidi is seeing is actually in the file and not something random happening elsewhere). It would appear some editors and players chop off the end of a MIDI file where they feel the data isn't needed (ie: end at the last event that ends any active notes, there are a couple, as no more data is required)
[Answer from Eric] I've run into this problem some too, and better yet, I've even caused the problem myself, so I think I may know at least one thing that could be causing it. When I was writing the MIDI file output code for the MOD->MIDI file conversion routines in TiMidity++, I had a few screwups calculating/writing the delta-times for the events. What would happen is that some event somewhere in the midi would have its delta time value (execute event at delta time since previous event on that channel) set to some huge incorrect value due to some stupid bug in my routines. When the midi gets to what would normally be the end of the MIDI file, it continues to play silence until the event with the bogus delta time finally gets executed at the time the midi told it to execute, which may be minutes or hours later depending on the magnitude of the screwup. Another cause may have been when I was miscalulating the size of the track to write into the header, so some of the last few events never got read in (since they were beyond the stated length of the track). I have an old DOS command line tool called MIDIFIX that does an excellent job of salvaging screwed up midi files. It told me exactly what was wrong with the "hung" midi files and was able to easily fix them (at least I think I remember it being able to handle them). It can deal with missing or truncated tracks too. I think it was part of an archive called MIDTOOLS? I can look for it later. I believe it is free software. No src, so for Linux you'd need to run it under DOSEmu. (Update) midifix can be found here: http://www.gnmidi.com/gnfreeen.htm
Retrieved from "http://alsa.opensrc.org/WildMidi"