diff options
| -rw-r--r-- | README.md | 95 | ||||
| -rw-r--r-- | broadcast.py | 62 | ||||
| -rw-r--r-- | voice.py | 6 |
3 files changed, 127 insertions, 36 deletions
@@ -1,29 +1,96 @@ # What this is -The ITL Chorus is a very simple, loosely bundled package for playing MIDI in real time over a network. Presently, it consists of three different frontends: +The ITL Chorus is a very simple, loosely bundled package for playing MIDI in +real time over a network. Presently, it consists of three different frontends: -- `mkiv.py`: Makes an *i*nter*v*al (`.iv`) file from a MIDI (usually `.mid`) file. The interval file is an XML document (easily compressed) consisting of the necessary information required to play back *voices* or *streams* such that no notes overlap (and duration information is available). -- `client.c`: A bare-minimum C program designed to run on even the most spartan Linux systems; it basically implements `beep` over UDP. -- `broadcast.py`: Accepts an interval file, assigns clients to streams, and plays a piece in real time. +- `mkiv.py`: Makes an *i*nter*v*al (`.iv`) file from a MIDI (usually `.mid`) + file. The interval file is an XML document (easily compressed) consisting of + the necessary information required to play back *voices* or *streams* such + that no notes overlap (and duration information is available). +- `client.c`: A bare-minimum C program designed to run on even the most spartan + Linux systems; it basically implements `beep` over UDP. +- `client.py`: A far-more-functional Python program with advanced options, + using the pyaudio API. +- `broadcast.py`: Accepts an interval file, assigns clients to streams, and + plays a piece in real time. -In general, you would use the tooling in precisely this order; generate an interval file with `mkiv.py` from a good MIDI performance (of your own acquiry :), compile and run `./client` *as root* on all the machines on a LAN that you would like to beep along, and then run `broadcast.py` with the generated interval file on any machine also on that LAN (potentially also one of the clients). +In general, you would use the tooling in precisely this order; generate an +interval file with `mkiv.py` from a good MIDI performance (of your own acquiry +:), either compile and run `./client` *as root* or run python client.py on all +the machines on a LAN that you would like to beep along, and then run +`broadcast.py` with the generated interval file on any machine also on that LAN +(potentially also one of the clients). # Troubleshooting In my experience, the most annoying errors come about as the following: -- No PC speaker. Many modern computers/motherboards omit this ancient piece of IBM technology entirely. Presumably, emulation is available (especially in desktop environments), but this normally requires a kernel-mode driver (as it has to respond to the very low-level syscall that actually would beep the speaker). ALSA purportedly provides snd_pcsp, but I've not seen it work yet. -- Network issues. `client` doesn't really check for any LAN, happily listening on whatever interfaces it can find at the time. Many very basic installations of Linux seem to not `dhclient` properly, even if the link is up, so you will want to make sure that your ip information is set up how you like it *before* running `client`. -- Lack of a compiler. Again, some bare-bones distributions don't ship with a compiler by default (I don't even know how you can use Linux like that :). Many nonetheless have package managers that will get a compiler and build environment for you. (On Debian and derivatives, `build-essential` works.) +- No PC speaker. Many modern computers/motherboards omit this ancient piece of + IBM technology entirely. Presumably, emulation is available (especially in + desktop environments), but this normally requires a kernel-mode driver (as it + has to respond to the very low-level syscall that actually would beep the + speaker). ALSA purportedly provides snd_pcsp, but I've not seen it work yet. + It should be noted that the python client.py script uses *regular* PCM + speakers to operate, and so can work under these conditions. +- Network issues. `client` doesn't really check for any LAN, happily listening + on whatever interfaces it can find at the time. Many very basic installations + of Linux seem to not `dhclient` properly, even if the link is up, so you will + want to make sure that your ip information is set up how you like it *before* + running `client`. +- Lack of a compiler. Again, some bare-bones distributions don't ship with a + compiler by default (I don't even know how you can use Linux like that :). + Many nonetheless have package managers that will get a compiler and build + environment for you. (On Debian and derivatives, `build-essential` works.) Please submit an issue if something else seems off! -# Obscure features +# Options -Not particularly well documented, `broadcast` supports some silly command-line magic: +All the scripts here (except the C program) have a plethora of options, most +of which are documented both at the beginning of the source and if you simply +pass `--help` or `-h` to them. Feel free to experiment! -- Using `-q` as a filename sends QUIT to all reachable clients, causing them to exit. -- Using `-t` as a filename sends test tones (440 for 0.25s, 880 for 0.25s) to all clients. The latter tone overlaps with the former tone, so you should hear N+1 tones (with N-1 octave chords) play across all clients in a somewhat non-deterministic order. From this, it should be easy to infer if a client is not network-reachable, or has a bad speaker. -- When playing a file, if a floating-point value is specified as the second argument, this represents a time remapping. If the original MIDI was consistently too slow or too fast, you can use values less than or greater than 1, respectively, to control how real time is mapped to stream time. +# Hacking -Note that `-q` and `-t` "as a filename" really means "as a filename"; `broadcast` will usually report a "Bad file" error afterward as it tries to open files named `-q` and `-t` after doing the relevant special command. (Alternatively, if they do exist and are valid interval files, it will play them :) +The .iv file format that is used extensively in communicating information is +certainly not any standard that I am aware of, but it seems like a rather +convenient standard for simple authorship. While I have no plans to write an +IV editor at the moment, that may change; in the meantime, whosoever would like +to do so should know the following about the IV files: + +- They are in XML--go ahead and open them in your favorite XML browser! +- Their root element is "iv" with no decided namespace yet. +- Under the root, there is a "meta" element with metainformation about the + compilation process--at present, this includes things like the "bpms" element + which has a "bpm" for each time period parsed out of the original MIDI. +- Also under the root, and arguably most importantly, is the "streams" element + that possesses all the "stream" elements that correspond to playable voices. +- Each stream has a "type" attribute which determines what it is ("ns" is a note + stream--the ones that have playable notes, and "aux" is a stream of non-note + MIDI events), and an optional "group" attribute which determines what group + it belongs to (`broadcast.py` uses this for routing). +- All note streams (type="ns") *should* contain non-overlapping notes, in the + sense that any "note" element in there should have a `time + dur` not greater + than its next note. Additionally, all notes in such a stream *should* be + sorted by time. Breaking either of these standards is not an egregious violation, + and may prove to be interesting, but (at the moment) it will prevent `broadcast.py` + from working properly. In addition, it should be noted that the clients are + *designed* to overwrite one incoming note with the next, regardless of whether or + not this interrupts the duration of the previous one--this is how "live mode" and "silence" work. + +# Todo + +- Polyphony--have multiple voices on one machine + - Mixed polyphony: the audio is the result of mixing (saturating addition, etc.)--only doable with PCM + - LFO polyphony: tones are "rapidly" switched between (how old microcomputers used to accomplish this with one beep speaker) +- Preloading--send events to clients early to avoid jitter problems with the network + - Would require a network time synchronization to work effectively; makes the broadcaster have less control over nuanced timing +- Other stream types--e.g., PCM streams for raw audio data + - More clientside implementation work + - Definitely a higher bandwidth--might interfere with critical timing, and would almost certainly need preloading +- Percussion--implement percussive instruments + - Requires an analysis of how current DAWs and MIDI editors mark "percussion" tracks--with a program change? GM specifies channel 10... +- Soundfonts--have the ability to significantly affect the instrumentation of the clients + - Would also be nice to do this from the broadcaster's end without introducing RCE + - Might require integration of another large libary like fluidsynth--at which point this would just be "networked MIDI" :) +- Code cleanup--make the entire project slightly more modular and palatable diff --git a/broadcast.py b/broadcast.py index 5606137..3b163b1 100644 --- a/broadcast.py +++ b/broadcast.py @@ -24,12 +24,13 @@ parser.add_option('-P', '--play-async', dest='play_async', action='store_true', parser.add_option('-D', '--duration', dest='duration', type='float', help='How long to play this note for') parser.add_option('-V', '--volume', dest='volume', type='int', help='Master volume (0-255)') parser.add_option('-s', '--silence', dest='silence', action='store_true', help='Instruct all clients to stop playing any active tones') +parser.add_option('-S', '--seek', dest='seek', type='float', help='Start time in seconds (scaled by --factor)') parser.add_option('-f', '--factor', dest='factor', type='float', help='Rescale time by this factor (0<f<1 are faster; 0.5 is twice the speed, 2 is half)') parser.add_option('-r', '--route', dest='routes', action='append', help='Add a routing directive (see --route-help)') parser.add_option('-v', '--verbose', dest='verbose', action='store_true', help='Be verbose; dump events and actual time (can slow down performance!)') parser.add_option('-W', '--wait-time', dest='wait_time', type='float', help='How long to wait for clients to initially respond (delays all broadcasts)') parser.add_option('--help-routes', dest='help_routes', action='store_true', help='Show help about routing directives') -parser.set_defaults(routes=[], random=0.0, rand_low=80, rand_high=2000, live=None, factor=1.0, duration=1.0, volume=255, wait_time=0.25, play=[], transpose=0) +parser.set_defaults(routes=[], random=0.0, rand_low=80, rand_high=2000, live=None, factor=1.0, duration=1.0, volume=255, wait_time=0.25, play=[], transpose=0, seek=0.0) options, args = parser.parse_args() if options.help_routes: @@ -70,6 +71,8 @@ try: except socket.timeout: pass +print len(clients), 'detected clients' + print 'Clients:' for cl in clients: print cl, @@ -86,10 +89,10 @@ for cl in clients: uid_groups.setdefault(uid, []).append(cl) type_groups.setdefault(tp, []).append(cl) if options.test: - s.sendto(str(Packet(CMD.PLAY, 0, 250000, 440, 255)), cl) + s.sendto(str(Packet(CMD.PLAY, 0, 250000, 440, options.volume)), cl) if not options.sync_test: time.sleep(0.25) - s.sendto(str(Packet(CMD.PLAY, 0, 250000, 880, 255)), cl) + s.sendto(str(Packet(CMD.PLAY, 0, 250000, 880, options.volume)), cl) if options.quit: s.sendto(str(Packet(CMD.QUIT)), cl) if options.silence: @@ -132,7 +135,7 @@ if options.live or options.list_live: exit() seq = sequencer.SequencerRead(sequencer_resolution=120) client_set = set(clients) - active_set = {} # note (pitch) -> client + active_set = {} # note (pitch) -> [client] deferred_set = set() # pitches held due to sustain sustain_status = False client, _, port = options.live.partition(',') @@ -154,7 +157,7 @@ if options.live or options.list_live: elif ev.type == S.SND_SEQ_EVENT_CONTROLLER: event = midi.ControlChangeEvent(channel = ev.data.control.channel, control = ev.data.control.param, value = ev.data.control.value) elif ev.type == S.SND_SEQ_EVENT_PGMCHANGE: - event = midi.ProgramChangeEvent(channel = ev.data.control.channel, pitch = ev.data.control.value) + event = midi.ProgramChangeEvent(channel = ev.data.control.channel, value = ev.data.control.value) elif ev.type == S.SND_SEQ_EVENT_PITCHBEND: event = midi.PitchWheelEvent(channel = ev.data.control.channel, pitch = ev.data.control.value) elif options.verbose: @@ -169,34 +172,36 @@ if options.live or options.list_live: if event.pitch in active_set: if sustain_status: deferred_set.discard(event.pitch) - else: - print 'WARNING: Note already activated: %r'%(event.pitch,), - continue - inactive_set = client_set - set(active_set.values()) + inactive_set = client_set - set(sum(active_set.values(), [])) if not inactive_set: print 'WARNING: Out of clients to do note %r; dropped'%(event.pitch,) continue - cli = random.choice(list(inactive_set)) + cli = sorted(inactive_set)[0] s.sendto(str(Packet(CMD.PLAY, 65535, 0, int(440.0 * 2**((event.pitch-69)/12.0)), 2*event.velocity)), cli) - active_set[event.pitch] = cli + active_set.setdefault(event.pitch, []).append(cli) + if options.verbose: + print 'LIVE:', event.pitch, '+ =>', active_set[event.pitch] elif isinstance(event, midi.NoteOffEvent): - if event.pitch not in active_set: + if event.pitch not in active_set or not active_set[event.pitch]: print 'WARNING: Deactivating inactive note %r'%(event.pitch,) continue if sustain_status: deferred_set.add(event.pitch) continue - s.sendto(str(Packet(CMD.PLAY, 0, 1, 1, 0)), active_set[event.pitch]) - del active_set[event.pitch] + cli = active_set[event.pitch].pop() + s.sendto(str(Packet(CMD.PLAY, 0, 1, 1, 0)), cli) + if options.verbose: + print 'LIVE:', event.pitch, '- =>', active_set[event.pitch] elif isinstance(event, midi.ControlChangeEvent): if event.control == 64: sustain_status = (event.value >= 64) if not sustain_status: for pitch in deferred_set: - if pitch not in active_set: + if pitch not in active_set or not active_set[pitch]: print 'WARNING: Attempted deferred removal of inactive note %r'%(pitch,) continue - s.sendto(str(Packet(CMD.PLAY, 0, 1, 1, 0)), active_set[pitch]) + for cli in active_set[pitch]: + s.sendto(str(Packet(CMD.PLAY, 0, 1, 1, 0)), cli) del active_set[pitch] deferred_set.clear() @@ -315,7 +320,20 @@ for fname in args: for route in routeset.routes: print route +<<<<<<< HEAD +class NSThread(threading.Thread): + def drop_missed(self): + nsq, cl = self._Thread__args + cnt = 0 + while nsq and float(nsq[0].get('time'))*factor < time.time() - BASETIME: + nsq.pop(0) + cnt += 1 + if options.verbose: + print self, 'dropped', cnt, 'notes due to miss' + self._Thread__args = (nsq, cl) +======= class NSThread(threading.Thread): +>>>>>>> 7c9661d892f6145d123d91924b720d9d87b69502 def wait_for(self, t): if t <= 0: return @@ -347,6 +365,17 @@ for fname in args: print 'Playback threads:' for thr in threads: print thr._Thread__args[1] +<<<<<<< HEAD + + BASETIME = time.time() - (options.seek*factor) + if options.seek > 0: + for thr in threads: + thr.drop_missed() + for thr in threads: + thr.start() + for thr in threads: + thr.join() +======= BASETIME = time.time() for thr in threads: @@ -354,4 +383,5 @@ for fname in args: for thr in threads: thr.join() +>>>>>>> 7c9661d892f6145d123d91924b720d9d87b69502 print fname, ': Done!' @@ -114,9 +114,3 @@ class VSumMixer(Voice): self.voices = list(voices) def __call__(self, theta): return norm_amp(sum([i(theta) for i in self.voices])) - -class object(object): - def __init__(self): - this_obj = object() - -foo = object() |