Replace ad-hoc subprocess with python-sounddevice stream callbacks in TTS

[PeterBowman]

In #32, I reworked the microphone data acquisition pipeline for our ASR app by using a simple raw input stream callback managed by PortAudio (through the Python sounddevice package, see sounddevice.RawInputStream):

speech/programs/speechRecognition/speechRecognition.py

Lines 263 to 283 in 349a097

	with sd.RawInputStream(blocksize=int(2880 / 2), # FIXME: hardcoded for pocketpshinx, vosk used to have 8000 here
	device=args.device,
	dtype='int16',
	channels=1,
	callback=lambda indata, frames, time, status: q.put(bytes(indata))) as stream:
	responder = responder_factory.create(stream)
	responder.yarp().attachAsServer(configPort)
	while True:
	frame = q.get()
	isPartial, transcription = responder.transcribe(frame)
	if transcription:
	if not isPartial:
	print('result: %s' % transcription)
	b = asrPort.prepare()
	b.clear()
	b.addString(transcription)
	asrPort.write()
	else:
	print('partial: %s' % transcription)

I presume a similar data flow could be implemented in the TTS app with sounddevice.RawOutputStream, thus replacing the ad-hoc subprocess and temporary file hacks currently used:

speech/programs/speechSynthesis/speechSynthesis.py

Line 19 in 349a097

PLAY_PROGRAMS = ['paplay', 'play -q', 'aplay -q']

speech/programs/speechSynthesis/speechSynthesis.py

Lines 102 to 122 in 349a097

	with tempfile.NamedTemporaryFile(mode='wb+', suffix='.wav') as wav_file:
	wav_file.write(wav_bytes)
	wav_file.seek(0)
	for play_program in reversed(PLAY_PROGRAMS):
	play_cmd = shlex.split(play_program)
	if not shutil.which(play_cmd[0]):
	continue
	play_cmd.append(wav_file.name)
	self.is_playing = True
	with subprocess.Popen(play_cmd) as self.p:
	try:
	self.p.wait()
	except: # e.g. on keyboard interrupt
	self.p.kill()
	self.is_playing = False
	break

This would also pave the way for the migration from Mimic 3 to Piper, see #30 (comment).

[PeterBowman]

Done at c74f814 (I seized the opportunity to migrate to Piper).