async def async_test_session(self):
"""Check good start/stop session."""
fake_transcription = Transcription(
text="this is a test", likelihood=1, transcribe_seconds=0, wav_seconds=0
)
def fake_transcribe(stream, *args):
"""Return test trancription."""
for chunk in stream:
if not chunk:
break
return fake_transcription
self.transcriber.transcribe_stream = fake_transcribe
# Start session
start_listening = AsrStartListening(
site_id=self.site_id,
session_id=self.session_id,
stop_on_silence=False,
send_audio_captured=True,
)
result = None
async for response in self.hermes.on_message_blocking(start_listening):
result = response
# No response expected
self.assertIsNone(result)
# Send in "audio"
fake_wav_bytes = self.hermes.to_wav_bytes(secrets.token_bytes(100))
fake_frame = AudioFrame(wav_bytes=fake_wav_bytes)
async for response in self.hermes.on_message_blocking(
fake_frame, site_id=self.site_id
):
result = response
# No response expected
self.assertIsNone(result)
# Stop session
stop_listening = AsrStopListening(
site_id=self.site_id, session_id=self.session_id
)
results = []
async for response in self.hermes.on_message_blocking(stop_listening):
results.append(response)
# Check results
self.assertEqual(
results,
[
AsrRecordingFinished(site_id=self.site_id, session_id=self.session_id),
AsrTextCaptured(
text=fake_transcription.text,
likelihood=fake_transcription.likelihood,
seconds=fake_transcription.transcribe_seconds,
site_id=self.site_id,
session_id=self.session_id,
),
(
AsrAudioCaptured(wav_bytes=fake_wav_bytes),
{"site_id": self.site_id, "session_id": self.session_id},
),
],
)
def publish_chunks(self):
"""Publish audio chunks to MQTT or UDP."""
try:
udp_dest = (self.udp_audio_host, self.udp_audio_port)
while not self._exit_requested:
chunk = self.chunk_queue.get()
if chunk:
# MQTT output
with io.BytesIO() as wav_buffer:
wav_file: wave.Wave_write = wave.open(wav_buffer, "wb")
with wav_file:
wav_file.setframerate(self.sample_rate)
wav_file.setsampwidth(self.sample_width)
wav_file.setnchannels(self.channels)
wav_file.writeframes(chunk)
wav_bytes = wav_buffer.getvalue()
if self.udp_output:
# UDP output
self.udp_socket.sendto(wav_bytes, udp_dest)
else:
# Publish to output site_id
self.publish(
AudioFrame(wav_bytes=wav_bytes),
site_id=self.output_site_id,
)
if self._dump_file is not None:
# print("tell is", self._dump_file.tell(), end=' ')
# write_wave( self._dump_file, wav_bytes, remove_header=True)
if USE_SOUNDFILE:
# soultion soundfile
self._dump_file.write(
np.frombuffer(
wav_bytes[44:],
np.int16)) # removing header!
else:
# Solution wave, this write always the latest buffer and doesnt happend it!
self._dump_file.writeframesraw(
wav_bytes[44:]) # removing header!
if self.enable_summary:
self.summary_frames_left -= 1
if self.summary_frames_left > 0:
continue
self.summary_frames_left = self.summary_skip_frames
if not self.vad:
# Create voice activity detector
self.vad = webrtcvad.Vad()
self.vad.set_mode(self.vad_mode)
# webrtcvad needs 16-bit 16Khz mono
# TODO: would be possible to split here if demux is not selected? this would avoid resampling,
# which is called continuously. (uncomment this code). With the switch --demux a proper channel
# is produced
# with io.BytesIO(wav_bytes) as wav_io:
# with wave.open(wav_io, "rb") as wav_file:
# if (wav_file.getframerate() != 16000) or \
# (wav_file.getsampwidth() != 2) or \
# (wav_file.getnchannels() != 1):
# print("Need Resample: sr={}, width={}, n_ch={}".format(wav_file.getframerate(),
# wav_file.getsampwidth(),
# wav_file.getnchannels()))
# else:
# print("No resample")
# webrtcvad needs 16-bit 16Khz mono
self.vad_audio_data += self.maybe_convert_wav(
wav_bytes,
sample_rate=16000,
sample_width=2,
channels=1)
is_speech = False
# Process in chunks of 30ms for webrtcvad
while len(self.vad_audio_data) >= self.vad_chunk_size:
vad_chunk = self.vad_audio_data[:self.
vad_chunk_size]
self.vad_audio_data = self.vad_audio_data[
self.vad_chunk_size:]
# Speech in any chunk counts as speech
is_speech = is_speech or self.vad.is_speech(
vad_chunk, 16000)
# Publish audio summary
self.publish(
AudioSummary(
debiased_energy=AudioSummary.
get_debiased_energy(chunk),
is_speech=is_speech,
),
site_id=self.output_site_id,
)
except Exception as e:
_LOGGER.exception("publish_chunks")
self.publish(
AudioRecordError(error=str(e),
context="publish_chunks",
site_id=self.site_id))
def publish_chunks(self):
"""Publish audio chunks to MQTT or UDP."""
try:
udp_dest = (self.udp_audio_host, self.udp_audio_port)
while True:
chunk = self.chunk_queue.get()
if chunk:
if self.test_audio_buffer:
# Add to buffer for microphone test
self.test_audio_buffer += chunk
# MQTT output
with io.BytesIO() as wav_buffer:
wav_file: wave.Wave_write = wave.open(wav_buffer, "wb")
with wav_file:
wav_file.setframerate(self.sample_rate)
wav_file.setsampwidth(self.sample_width)
wav_file.setnchannels(self.channels)
wav_file.writeframes(chunk)
wav_bytes = wav_buffer.getvalue()
if self.udp_output:
# UDP output
self.udp_socket.sendto(wav_bytes, udp_dest)
else:
# Publish to output site_id
self.publish(
AudioFrame(wav_bytes=wav_bytes),
site_id=self.output_site_id,
)
if self.enable_summary:
self.summary_frames_left -= 1
if self.summary_frames_left > 0:
continue
self.summary_frames_left = self.summary_skip_frames
if not self.vad:
# Create voice activity detector
self.vad = webrtcvad.Vad()
self.vad.set_mode(self.vad_mode)
# webrtcvad needs 16-bit 16Khz mono
self.vad_audio_data += self.maybe_convert_wav(
wav_bytes, sample_rate=16000, sample_width=2, channels=1
)
is_speech = False
# Process in chunks of 30ms for webrtcvad
while len(self.vad_audio_data) >= self.vad_chunk_size:
vad_chunk = self.vad_audio_data[: self.vad_chunk_size]
self.vad_audio_data = self.vad_audio_data[
self.vad_chunk_size :
]
# Speech in any chunk counts as speech
is_speech = is_speech or self.vad.is_speech(
vad_chunk, 16000
)
# Publish audio summary
self.publish(
AudioSummary(
debiased_energy=AudioSummary.get_debiased_energy(chunk),
is_speech=is_speech,
),
site_id=self.output_site_id,
)
except Exception as e:
_LOGGER.exception("publish_chunks")
self.publish(
AudioRecordError(
error=str(e), context="publish_chunks", site_id=self.output_site_id
)
)
