Support for polyphonic synthetic sounds in sound plugin
This commit is contained in:
parent
7a5657778e
commit
c2cf0b5f74
1 changed files with 148 additions and 112 deletions
|
@ -70,9 +70,6 @@ class Sound(object):
|
||||||
raise RuntimeError('Please specify either a MIDI note or a base ' +
|
raise RuntimeError('Please specify either a MIDI note or a base ' +
|
||||||
'frequency')
|
'frequency')
|
||||||
|
|
||||||
# TODO support for multiple notes/frequencies, either for chords or
|
|
||||||
# harmonics
|
|
||||||
|
|
||||||
if midi_note:
|
if midi_note:
|
||||||
self.midi_note = midi_note
|
self.midi_note = midi_note
|
||||||
self.frequency = self.note_to_freq(midi_note=midi_note,
|
self.frequency = self.note_to_freq(midi_note=midi_note,
|
||||||
|
@ -151,6 +148,25 @@ class Sound(object):
|
||||||
'duration': duration,
|
'duration': duration,
|
||||||
})
|
})
|
||||||
|
|
||||||
|
@classmethod
|
||||||
|
def build(cls, *args, **kwargs):
|
||||||
|
"""
|
||||||
|
Construct a sound object either from a JSON representation or a
|
||||||
|
key-value representation
|
||||||
|
"""
|
||||||
|
|
||||||
|
if args:
|
||||||
|
if isinstance(args[0], cls):
|
||||||
|
return args[0]
|
||||||
|
if isinstance(args[0], str):
|
||||||
|
kwargs = json.loads(args[0])
|
||||||
|
elif isinstance(args[0], dict):
|
||||||
|
kwargs = args[0]
|
||||||
|
if kwargs:
|
||||||
|
return Sound(**kwargs)
|
||||||
|
|
||||||
|
raise RuntimeError('Usage: {}'.format(__doc__))
|
||||||
|
|
||||||
|
|
||||||
class SoundPlugin(Plugin):
|
class SoundPlugin(Plugin):
|
||||||
"""
|
"""
|
||||||
|
@ -198,6 +214,8 @@ class SoundPlugin(Plugin):
|
||||||
self.recording_state = RecordingState.STOPPED
|
self.recording_state = RecordingState.STOPPED
|
||||||
self.recording_state_lock = RLock()
|
self.recording_state_lock = RLock()
|
||||||
self.recording_paused_changed = Event()
|
self.recording_paused_changed = Event()
|
||||||
|
self.active_players = {}
|
||||||
|
self.completed_callback_events = {}
|
||||||
|
|
||||||
def _get_default_device(self, category):
|
def _get_default_device(self, category):
|
||||||
"""
|
"""
|
||||||
|
@ -257,9 +275,43 @@ class SoundPlugin(Plugin):
|
||||||
|
|
||||||
return devs
|
return devs
|
||||||
|
|
||||||
|
def _play_audio_callback(self, q, blocksize, streamtype):
|
||||||
|
import sounddevice as sd
|
||||||
|
|
||||||
|
is_raw_stream = streamtype == sd.RawOutputStream
|
||||||
|
def audio_callback(outdata, frames, time, status):
|
||||||
|
if self._get_playback_state() == PlaybackState.STOPPED:
|
||||||
|
raise sd.CallbackAbort
|
||||||
|
|
||||||
|
while self._get_playback_state() == PlaybackState.PAUSED:
|
||||||
|
self.playback_paused_changed.wait()
|
||||||
|
|
||||||
|
assert frames == blocksize
|
||||||
|
if status.output_underflow:
|
||||||
|
self.logger.warning('Output underflow: increase blocksize?')
|
||||||
|
outdata = (b'\x00' if is_raw_stream else 0.) * len(outdata)
|
||||||
|
return
|
||||||
|
|
||||||
|
assert not status
|
||||||
|
|
||||||
|
try:
|
||||||
|
data = q.get_nowait()
|
||||||
|
except queue.Empty:
|
||||||
|
self.logger.warning('Buffer is empty: increase buffersize?')
|
||||||
|
raise sd.CallbackAbort
|
||||||
|
|
||||||
|
if len(data) < len(outdata):
|
||||||
|
outdata[:len(data)] = data
|
||||||
|
outdata[len(data):] = (b'\x00' if is_raw_stream else 0.) * \
|
||||||
|
(len(outdata) - len(data))
|
||||||
|
else:
|
||||||
|
outdata[:] = data
|
||||||
|
|
||||||
|
return audio_callback
|
||||||
|
|
||||||
|
|
||||||
@action
|
@action
|
||||||
def play(self, file=None, sounds=None, device=None, blocksize=None,
|
def play(self, file=None, sound=None, device=None, blocksize=None,
|
||||||
bufsize=Sound._DEFAULT_BUFSIZE, samplerate=None, channels=None):
|
bufsize=Sound._DEFAULT_BUFSIZE, samplerate=None, channels=None):
|
||||||
"""
|
"""
|
||||||
Plays a sound file (support formats: wav, raw) or a synthetic sound.
|
Plays a sound file (support formats: wav, raw) or a synthetic sound.
|
||||||
|
@ -267,20 +319,17 @@ class SoundPlugin(Plugin):
|
||||||
:param file: Sound file path. Specify this if you want to play a file
|
:param file: Sound file path. Specify this if you want to play a file
|
||||||
:type file: str
|
:type file: str
|
||||||
|
|
||||||
:param sounds: Sounds to play. Specify this if you want to play
|
:param sound: Sound to play. Specify this if you want to play
|
||||||
synthetic sounds. TODO: So far only one single-frequency sound is
|
synthetic sounds. You can also create polyphonic sounds by just
|
||||||
supported, support for multiple sounds, chords, harmonics and mixed
|
calling play multple times.
|
||||||
sounds is ont the way.
|
:type sound: Sound. You can initialize it either from a list
|
||||||
:type sounds: list[Sound]. You can initialize it either from a list
|
|
||||||
of `Sound` objects or from its JSON representation, e.g.:
|
of `Sound` objects or from its JSON representation, e.g.:
|
||||||
|
|
||||||
[
|
{
|
||||||
{
|
"midi_note": 69, # 440 Hz A
|
||||||
"midi_note": 69, # 440 Hz A
|
"gain": 1.0, # Maximum volume
|
||||||
"gain": 1.0, # Maximum volume
|
"duration": 1.0 # 1 second or until release/pause/stop
|
||||||
"duration": 1.0 # 1 second or until release/pause/stop
|
}
|
||||||
}
|
|
||||||
]
|
|
||||||
|
|
||||||
:param device: Output device (default: default configured device or
|
:param device: Output device (default: default configured device or
|
||||||
system default audio output if not configured)
|
system default audio output if not configured)
|
||||||
|
@ -302,24 +351,18 @@ class SoundPlugin(Plugin):
|
||||||
:type channels: int
|
:type channels: int
|
||||||
"""
|
"""
|
||||||
|
|
||||||
if not file and not sounds:
|
if not file and not sound:
|
||||||
raise RuntimeError('Please specify either a file to play or a ' +
|
raise RuntimeError('Please specify either a file to play or a ' +
|
||||||
'list of sound objects')
|
'list of sound objects')
|
||||||
|
|
||||||
import sounddevice as sd
|
import sounddevice as sd
|
||||||
|
|
||||||
if self._get_playback_state() != PlaybackState.STOPPED:
|
|
||||||
if file:
|
|
||||||
self.logger.info('Stopping playback before playing')
|
|
||||||
self.stop_playback()
|
|
||||||
time.sleep(2)
|
|
||||||
|
|
||||||
if blocksize is None:
|
if blocksize is None:
|
||||||
blocksize = self.output_blocksize
|
blocksize = self.output_blocksize
|
||||||
|
|
||||||
self.playback_paused_changed.clear()
|
self.playback_paused_changed.clear()
|
||||||
|
|
||||||
completed_callback_event = Event()
|
stream_index = None
|
||||||
q = queue.Queue(maxsize=bufsize)
|
q = queue.Queue(maxsize=bufsize)
|
||||||
f = None
|
f = None
|
||||||
t = 0.
|
t = 0.
|
||||||
|
@ -332,60 +375,20 @@ class SoundPlugin(Plugin):
|
||||||
if device is None:
|
if device is None:
|
||||||
device = self._get_default_device('output')
|
device = self._get_default_device('output')
|
||||||
|
|
||||||
def audio_callback(outdata, frames, time, status):
|
|
||||||
if self._get_playback_state() == PlaybackState.STOPPED:
|
|
||||||
raise sd.CallbackAbort
|
|
||||||
|
|
||||||
while self._get_playback_state() == PlaybackState.PAUSED:
|
|
||||||
self.playback_paused_changed.wait()
|
|
||||||
|
|
||||||
assert frames == blocksize
|
|
||||||
if status.output_underflow:
|
|
||||||
self.logger.warning('Output underflow: increase blocksize?')
|
|
||||||
outdata = (b'\x00' if file else 0.) * len(outdata)
|
|
||||||
return
|
|
||||||
|
|
||||||
assert not status
|
|
||||||
|
|
||||||
try:
|
|
||||||
data = q.get_nowait()
|
|
||||||
except queue.Empty:
|
|
||||||
self.logger.warning('Buffer is empty: increase buffersize?')
|
|
||||||
raise sd.CallbackAbort
|
|
||||||
|
|
||||||
if len(data) < len(outdata):
|
|
||||||
outdata[:len(data)] = data
|
|
||||||
outdata[len(data):] = (b'\x00' if file else 0.) * \
|
|
||||||
(len(outdata) - len(data))
|
|
||||||
|
|
||||||
# if f:
|
|
||||||
# raise sd.CallbackStop
|
|
||||||
else:
|
|
||||||
outdata[:] = data
|
|
||||||
|
|
||||||
|
|
||||||
try:
|
try:
|
||||||
if file:
|
if file:
|
||||||
import soundfile as sf
|
import soundfile as sf
|
||||||
f = sf.SoundFile(file)
|
f = sf.SoundFile(file)
|
||||||
|
|
||||||
if not samplerate:
|
if not samplerate:
|
||||||
samplerate = f.samplerate if f else Sound._DEFAULT_SAMPLERATE
|
samplerate = f.samplerate if f else Sound._DEFAULT_SAMPLERATE
|
||||||
|
|
||||||
if not channels:
|
if not channels:
|
||||||
channels = f.channels if f else 1
|
channels = f.channels if f else 1
|
||||||
|
|
||||||
self.start_playback()
|
self.logger.info('Starting playback of {} to device [{}]'.
|
||||||
self.logger.info('Started playback of {} to device [{}]'.
|
format(file or sound, device))
|
||||||
format(file or sounds, device))
|
|
||||||
|
|
||||||
if sounds:
|
if sound:
|
||||||
if isinstance(sounds, str):
|
sound = Sound.build(sound)
|
||||||
sounds = json.loads(sounds)
|
|
||||||
|
|
||||||
for i in range(0, len(sounds)):
|
|
||||||
if isinstance(sounds[i], dict):
|
|
||||||
sounds[i] = Sound(**(sounds[i]))
|
|
||||||
|
|
||||||
# Audio queue pre-fill loop
|
# Audio queue pre-fill loop
|
||||||
for _ in range(bufsize):
|
for _ in range(bufsize):
|
||||||
|
@ -394,8 +397,6 @@ class SoundPlugin(Plugin):
|
||||||
if not data:
|
if not data:
|
||||||
break
|
break
|
||||||
else:
|
else:
|
||||||
# TODO support for multiple sounds or mixed sounds
|
|
||||||
sound = sounds[0]
|
|
||||||
blocktime = float(blocksize / samplerate)
|
blocktime = float(blocksize / samplerate)
|
||||||
next_t = min(t+blocktime, sound.duration) \
|
next_t = min(t+blocktime, sound.duration) \
|
||||||
if sound.duration is not None else t+blocktime
|
if sound.duration is not None else t+blocktime
|
||||||
|
@ -410,17 +411,19 @@ class SoundPlugin(Plugin):
|
||||||
while self._get_playback_state() == PlaybackState.PAUSED:
|
while self._get_playback_state() == PlaybackState.PAUSED:
|
||||||
self.playback_paused_changed.wait()
|
self.playback_paused_changed.wait()
|
||||||
|
|
||||||
if self._get_playback_state() == PlaybackState.STOPPED:
|
|
||||||
raise sd.CallbackAbort
|
|
||||||
|
|
||||||
q.put_nowait(data) # Pre-fill the audio queue
|
q.put_nowait(data) # Pre-fill the audio queue
|
||||||
|
|
||||||
streamtype = sd.RawOutputStream if file else sd.OutputStream
|
streamtype = sd.RawOutputStream if file else sd.OutputStream
|
||||||
stream = streamtype(samplerate=samplerate, blocksize=blocksize,
|
stream = streamtype(samplerate=samplerate, blocksize=blocksize,
|
||||||
device=device, channels=channels,
|
device=device, channels=channels,
|
||||||
dtype='float32', callback=audio_callback,
|
dtype='float32',
|
||||||
|
callback=self._play_audio_callback(
|
||||||
|
q=q, blocksize=blocksize,
|
||||||
|
streamtype=streamtype),
|
||||||
finished_callback=completed_callback_event.set)
|
finished_callback=completed_callback_event.set)
|
||||||
|
|
||||||
|
stream_index = self.start_playback(stream, completed_callback_event)
|
||||||
|
|
||||||
with stream:
|
with stream:
|
||||||
# Timeout set until we expect all the buffered blocks to
|
# Timeout set until we expect all the buffered blocks to
|
||||||
# be consumed
|
# be consumed
|
||||||
|
@ -435,8 +438,6 @@ class SoundPlugin(Plugin):
|
||||||
if not data:
|
if not data:
|
||||||
break
|
break
|
||||||
else:
|
else:
|
||||||
# TODO support for multiple sounds or mixed sounds
|
|
||||||
sound = sounds[0]
|
|
||||||
blocktime = float(blocksize / samplerate)
|
blocktime = float(blocksize / samplerate)
|
||||||
next_t = min(t+blocktime, sound.duration) \
|
next_t = min(t+blocktime, sound.duration) \
|
||||||
if sound.duration is not None else t+blocktime
|
if sound.duration is not None else t+blocktime
|
||||||
|
@ -458,8 +459,6 @@ class SoundPlugin(Plugin):
|
||||||
raise e
|
raise e
|
||||||
|
|
||||||
completed_callback_event.wait()
|
completed_callback_event.wait()
|
||||||
# if sounds:
|
|
||||||
# sd.wait()
|
|
||||||
except queue.Full as e:
|
except queue.Full as e:
|
||||||
self.logger.warning('Playback timeout: audio callback failed?')
|
self.logger.warning('Playback timeout: audio callback failed?')
|
||||||
finally:
|
finally:
|
||||||
|
@ -467,7 +466,7 @@ class SoundPlugin(Plugin):
|
||||||
f.close()
|
f.close()
|
||||||
f = None
|
f = None
|
||||||
|
|
||||||
self.stop_playback()
|
self.stop_playback(stream_index)
|
||||||
|
|
||||||
|
|
||||||
@action
|
@action
|
||||||
|
@ -503,10 +502,6 @@ class SoundPlugin(Plugin):
|
||||||
|
|
||||||
import sounddevice as sd
|
import sounddevice as sd
|
||||||
|
|
||||||
if self._get_recording_state() != RecordingState.STOPPED:
|
|
||||||
self.stop_recording()
|
|
||||||
time.sleep(2)
|
|
||||||
|
|
||||||
self.recording_paused_changed.clear()
|
self.recording_paused_changed.clear()
|
||||||
|
|
||||||
if file:
|
if file:
|
||||||
|
@ -616,14 +611,6 @@ class SoundPlugin(Plugin):
|
||||||
|
|
||||||
import sounddevice as sd
|
import sounddevice as sd
|
||||||
|
|
||||||
if self._get_playback_state() != PlaybackState.STOPPED:
|
|
||||||
self.stop_playback()
|
|
||||||
time.sleep(2)
|
|
||||||
|
|
||||||
if self._get_recording_state() != RecordingState.STOPPED:
|
|
||||||
self.stop_recording()
|
|
||||||
time.sleep(2)
|
|
||||||
|
|
||||||
self.playback_paused_changed.clear()
|
self.playback_paused_changed.clear()
|
||||||
self.recording_paused_changed.clear()
|
self.recording_paused_changed.clear()
|
||||||
|
|
||||||
|
@ -655,46 +642,94 @@ class SoundPlugin(Plugin):
|
||||||
outdata[:] = indata
|
outdata[:] = indata
|
||||||
|
|
||||||
|
|
||||||
|
stream_index = None
|
||||||
|
|
||||||
try:
|
try:
|
||||||
import soundfile as sf
|
import soundfile as sf
|
||||||
import numpy
|
import numpy
|
||||||
|
|
||||||
with sd.Stream(samplerate=sample_rate, channels=channels,
|
stream = sd.Stream(samplerate=sample_rate, channels=channels,
|
||||||
blocksize=blocksize, latency=latency,
|
blocksize=blocksize, latency=latency,
|
||||||
device=(input_device, output_device),
|
device=(input_device, output_device),
|
||||||
dtype=dtype, callback=audio_callback):
|
dtype=dtype, callback=audio_callback)
|
||||||
self.start_recording()
|
self.start_recording()
|
||||||
self.start_playback()
|
stream_index = self.start_playback(stream)
|
||||||
|
|
||||||
self.logger.info('Started recording pass-through from device ' +
|
self.logger.info('Started recording pass-through from device ' +
|
||||||
'[{}] to device [{}]'.
|
'[{}] to device [{}]'.
|
||||||
format(input_device, output_device))
|
format(input_device, output_device))
|
||||||
|
|
||||||
recording_started_time = time.time()
|
recording_started_time = time.time()
|
||||||
|
|
||||||
while self._get_recording_state() != RecordingState.STOPPED \
|
while self._get_recording_state() != RecordingState.STOPPED \
|
||||||
and (duration is None or
|
and (duration is None or
|
||||||
time.time() - recording_started_time < duration):
|
time.time() - recording_started_time < duration):
|
||||||
while self._get_recording_state() == RecordingState.PAUSED:
|
while self._get_recording_state() == RecordingState.PAUSED:
|
||||||
self.recording_paused_changed.wait()
|
self.recording_paused_changed.wait()
|
||||||
|
|
||||||
time.sleep(0.1)
|
time.sleep(0.1)
|
||||||
|
|
||||||
except queue.Empty as e:
|
except queue.Empty as e:
|
||||||
self.logger.warning('Recording timeout: audio callback failed?')
|
self.logger.warning('Recording timeout: audio callback failed?')
|
||||||
finally:
|
finally:
|
||||||
self.stop_playback()
|
self.stop_playback(stream_index)
|
||||||
self.stop_recording()
|
self.stop_recording()
|
||||||
|
|
||||||
|
|
||||||
def start_playback(self):
|
@action
|
||||||
|
def get_active_players(self):
|
||||||
|
"""
|
||||||
|
:returns: A list of active players
|
||||||
|
"""
|
||||||
|
return {
|
||||||
|
i: {
|
||||||
|
attr: getattr(stream, attr)
|
||||||
|
for attr in ['active', 'closed', 'stopped', 'blocksize',
|
||||||
|
'channels', 'cpu_load', 'device', 'dtype',
|
||||||
|
'latency', 'samplerate', 'samplesize']
|
||||||
|
if hasattr(stream, attr)
|
||||||
|
} for i, stream in self.active_players.items()
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
def start_playback(self, stream, completed_callback_event=None):
|
||||||
|
stream_index = None
|
||||||
with self.playback_state_lock:
|
with self.playback_state_lock:
|
||||||
self.playback_state = PlaybackState.PLAYING
|
self.playback_state = PlaybackState.PLAYING
|
||||||
|
stream_index = len(self.active_players)
|
||||||
|
self.active_players[stream_index] = stream
|
||||||
|
self.completed_callback_events[stream_index] = \
|
||||||
|
completed_callback_event if completed_callback_event else Event()
|
||||||
|
|
||||||
|
self.logger.info('Playback started on stream index {}'.
|
||||||
|
format(stream_index))
|
||||||
|
return stream_index
|
||||||
|
|
||||||
@action
|
@action
|
||||||
def stop_playback(self):
|
def stop_playback(self, *streams):
|
||||||
with self.playback_state_lock:
|
with self.playback_state_lock:
|
||||||
self.playback_state = PlaybackState.STOPPED
|
if not streams:
|
||||||
|
streams = self.active_players.keys()
|
||||||
|
updated_n_players = len(self.active_players)
|
||||||
|
completed_callback_events = {}
|
||||||
|
|
||||||
|
for i in streams:
|
||||||
|
if i is None or not (i in self.active_players):
|
||||||
|
continue
|
||||||
|
|
||||||
|
stream = self.active_players[i]
|
||||||
|
updated_n_players -= 1
|
||||||
|
if self.completed_callback_events[i]:
|
||||||
|
completed_callback_events[i] = self.completed_callback_events[i]
|
||||||
|
|
||||||
|
if not updated_n_players:
|
||||||
|
self.playback_state = PlaybackState.STOPPED
|
||||||
|
|
||||||
|
for i, event in completed_callback_events.items():
|
||||||
|
event.wait()
|
||||||
|
del self.completed_callback_events[i]
|
||||||
|
del self.active_players[i]
|
||||||
|
|
||||||
self.logger.info('Playback stopped')
|
self.logger.info('Playback stopped')
|
||||||
|
|
||||||
@action
|
@action
|
||||||
|
@ -707,8 +742,9 @@ class SoundPlugin(Plugin):
|
||||||
else:
|
else:
|
||||||
return
|
return
|
||||||
|
|
||||||
self.logger.info('Playback paused state toggled')
|
|
||||||
self.playback_paused_changed.set()
|
self.playback_paused_changed.set()
|
||||||
|
self.logger.info('Playback ' + ('paused' if self.playback_state ==
|
||||||
|
PlaybackState.PAUSED else 'playing'))
|
||||||
|
|
||||||
def start_recording(self):
|
def start_recording(self):
|
||||||
with self.recording_state_lock:
|
with self.recording_state_lock:
|
||||||
|
|
Loading…
Reference in a new issue