def test_vad_is_triggered(mock_class):
context = SpeechContext()
detector = VoiceActivityDetector(sample_rate=16000,
frame_width=10,
vad_rise_delay=0,
vad_fall_delay=0)
frame = np.zeros(160, np.int16)
detector(context, frame)
assert context.is_speech
detector.close()
def test_vad_rise_delay():
context = SpeechContext()
detector = VoiceActivityDetector(frame_width=10, vad_rise_delay=30)
for i in range(3):
frame = voice_frame()
detector(context, frame)
if i < 2:
assert not context.is_speech
else:
assert context.is_speech
detector.close()
def test_vad_is_triggered():
context = SpeechContext()
detector = VoiceActivityDetector(frame_width=10)
frame = silence_frame()
detector(context, frame)
assert not context.is_speech
frame = voice_frame()
detector(context, frame)
assert context.is_speech
detector.close()
def test_vad_rise_delay(mock_class):
context = SpeechContext()
detector = VoiceActivityDetector(sample_rate=16000,
frame_width=10,
vad_rise_delay=30,
vad_fall_delay=0)
for i in range(3):
frame = np.zeros(160, np.int16)
detector(context, frame)
if i < 2:
assert not context.is_speech
else:
assert context.is_speech
detector.close()
def create(spokestack_id: str,
spokestack_secret: str,
sample_rate: int = 16000,
frame_width: int = 20,
**kwargs) -> SpeechPipeline:
"""
Args:
spokestack_id (str): spokestack API id.
spokestack_secret (str): Spokestack API secret.
sample_rate (int): sample rate of the audio (Hz).
frame_width (int): width of the audio frame: 10, 20, or 30 (ms).
Returns:
SpeechPipeline instance with profile configuration.
"""
pipeline = SpeechPipeline(
input_source=PyAudioInput(sample_rate=sample_rate,
frame_width=frame_width,
**kwargs),
stages=[
VoiceActivityDetector(sample_rate=sample_rate,
frame_width=frame_width,
**kwargs),
VoiceActivityTrigger(),
ActivationTimeout(frame_width=frame_width, **kwargs),
CloudSpeechRecognizer(spokestack_id=spokestack_id,
spokestack_secret=spokestack_secret,
sample_rate=sample_rate,
frame_width=frame_width,
**kwargs),
],
)
return pipeline
def test_invalid_dtype():
context = SpeechContext()
detector = VoiceActivityDetector()
bad_frame = np.random.rand(160)
with pytest.raises(Exception):
detector(context, bad_frame)
def test_vad_fall_untriggered():
context = SpeechContext()
detector = VoiceActivityDetector(frame_width=10, vad_fall_delay=20)
voice = voice_frame()
silence = silence_frame()
detector(context, voice)
assert context.is_speech
for i in range(10):
detector(context, silence)
assert context.is_speech
detector(context, silence)
assert not context.is_speech
detector.close()
def test_vad_fall_untriggered():
context = SpeechContext()
detector = VoiceActivityDetector(sample_rate=16000,
frame_width=10,
vad_rise_delay=0,
vad_fall_delay=20)
with patch("webrtcvad.Vad.is_speech", return_value=True):
frame = np.zeros(160, np.int16)
detector(context, frame)
assert context.is_speech
with patch("webrtcvad.Vad.is_speech", return_value=False):
frame = np.zeros(160, np.int16)
detector(context, frame)
assert context.is_speech
with patch("webrtcvad.Vad.is_speech", return_value=False):
frame = np.zeros(160, np.int16)
detector(context, frame)
assert not context.is_speech
detector.close()
def create(
spokestack_id: str,
spokestack_secret: str,
sample_rate: int = 16000,
frame_width: int = 20,
model_dir: str = "",
**kwargs: Any,
) -> SpeechPipeline:
"""Creates a speech pipeline instance from profile
Args:
spokestack_id (str): spokestack API id.
spokestack_secret (str): Spokestack API secret.
sample_rate (int): sample rate of the audio (Hz).
frame_width (int): width of the audio frame: 10, 20, or 30 (ms).
model_dir (str): Directory containing the tflite wakeword models.
Returns:
"""
pipeline = SpeechPipeline(
input_source=PyAudioInput(frame_width=frame_width,
sample_rate=sample_rate,
**kwargs),
stages=[
AutomaticGainControl(sample_rate=sample_rate,
frame_width=frame_width),
AutomaticNoiseSuppression(sample_rate=sample_rate),
VoiceActivityDetector(
frame_width=frame_width,
sample_rate=sample_rate,
**kwargs,
),
WakewordTrigger(model_dir=model_dir, **kwargs),
ActivationTimeout(frame_width=frame_width, **kwargs),
CloudSpeechRecognizer(
spokestack_secret=spokestack_secret,
spokestack_id=spokestack_id,
**kwargs,
),
],
)
return pipeline
def main():
pipeline = SpeechPipeline(
PyAudioInput(frame_width=20,
sample_rate=16000,
exception_on_overflow=False),
[
VoiceActivityDetector(),
WakewordTrigger(pre_emphasis=0.97, model_dir="tflite"),
GoogleSpeechRecognizer(GOOGLE_CREDS),
ActivationTimeout(),
],
)
dialogue_manager = DialogueManager(
"tflite", "distilbert-base-cased-distilled-squad")
manager = TextToSpeechManager(
TextToSpeechClient(KEY_ID, KEY_SECRET),
PyAudioOutput(),
)
@pipeline.event
def on_activate(context):
print(context.is_active)
@pipeline.event
def on_recognize(context):
pipeline.pause()
answer = dialogue_manager(context.transcript)
manager.synthesize(answer, "text", "demo-male")
pipeline.resume()
@pipeline.event
def on_deactivate(context):
print(context.is_active)
manager.synthesize(dialogue_manager.greet(), "text", "demo-male")
pipeline.start()
pipeline.run()
def main():
pipeline = SpeechPipeline(
PyAudioInput(frame_width=20, sample_rate=16000, exception_on_overflow=False),
[
VoiceActivityDetector(),
WakewordTrigger(pre_emphasis=0.97, model_dir="tflite"),
CloudSpeechRecognizer(spokestack_id=KEY_ID, spokestack_secret=KEY_SECRET),
ActivationTimeout(),
],
)
nlu = TFLiteNLU("tflite")
dialogue_manager = DialogueManager()
manager = TextToSpeechManager(
TextToSpeechClient(KEY_ID, KEY_SECRET), PyAudioOutput(),
)
@pipeline.event
def on_activate(context):
print("active")
@pipeline.event
def on_recognize(context):
pipeline.pause()
results = nlu(context.transcript)
response = dialogue_manager(results)
if response:
manager.synthesize(response, "text", "demo-male")
pipeline.resume()
if results.intent == "AMAZON.StopIntent":
pipeline.stop()
manager.synthesize(Response.WELCOME.value, "text", "demo-male")
pipeline.start()
pipeline.run()
def create(classes: List[str],
model_dir: str,
sample_rate: int = 16000,
frame_width: int = 20,
**kwargs: Any) -> SpeechPipeline:
"""Create a speech pipeline instance from profile.
Args:
model_dir (str): Directory containing the tflite keyword models.
classes: (List(str)): Classes for the keyword model to recognize
sample_rate (int): sample rate of the audio (Hz).
frame_width (int): width of the audio frame: 10, 20, or 30 (ms).
"""
pipeline = SpeechPipeline(
input_source=PyAudioInput(frame_width=frame_width,
sample_rate=sample_rate,
**kwargs),
stages=[
AutomaticGainControl(sample_rate=sample_rate,
frame_width=frame_width),
AutomaticNoiseSuppression(sample_rate=sample_rate),
VoiceActivityDetector(sample_rate=sample_rate,
frame_width=frame_width,
**kwargs),
VoiceActivityTrigger(),
KeywordRecognizer(
classes=classes,
model_dir=model_dir,
sample_rate=sample_rate,
**kwargs,
),
ActivationTimeout(frame_width=frame_width, **kwargs),
],
)
return pipeline
def test_invalid_sample_rate():
with pytest.raises(ValueError):
_ = VoiceActivityDetector(sample_rate=9000)
def test_invalid_frame_width():
with pytest.raises(ValueError):
_ = VoiceActivityDetector(frame_width=30)