def label_stream():
hotword_detected = False
countdown = 0
audio_stream = AudiostreamSource()
action_detector = AudioRecognition(libpath, action_graph, action_labels)
hotword_detector = AudioRecognition(libpath, hotword_graph, hotword_labels)
#
#action_detector = hotword_detector
hotword_detector.SetSensitivity(0.5)
action_detector.SetSensitivity(0.55)
bufsize = hotword_detector.GetInputDataSize()
audio_stream.start()
print("Audio Recognition Version: " + hotword_detector.GetVersionString())
try:
while (True):
frame = audio_stream.read(bufsize, bufsize)
if (not frame):
time.sleep(0.01)
continue
if (countdown > 0):
countdown -= 1
if (countdown == 0):
hotword_detected = False
print("Stopped Listening")
if (not hotword_detected):
prediction = hotword_detector.RunDetection(frame)
print(hotword_detector.GetPredictionLabel(prediction))
if (prediction
and hotword_detector.GetPredictionLabel(prediction)
== 'light'):
hotword_detected = True
countdown = 20
now = datetime.datetime.now().strftime("%d.%b %Y %H:%M:%S")
print("Listening")
else:
prediction = action_detector.RunDetection(frame)
if (prediction):
label = action_detector.GetPredictionLabel(prediction)
if (label == "on"):
print("Turning light on")
if (label == "off"):
print("Turning light off")
countdown = 0
hotword_detected = False
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)
def label_stream(labels, libpath, graph, sensitivity):
audio_stream = AudiostreamSource()
extractor = FeatureExtractor(libpath)
extactor_gain = 1.0
detector = AudioRecognition(libpath, graph, labels)
detector.SetSensitivity(sensitivity)
bufsize = detector.GetInputDataSize()
print("Audio Recognition Version: " + detector.GetVersionString())
audio_stream.start()
try:
while (True):
frame = audio_stream.read(bufsize * 2, bufsize * 2)
if (not frame):
time.sleep(0.01)
continue
features = extractor.signal_to_mel(frame, extactor_gain)
prediction = detector.RunDetection(features)
if (prediction):
now = datetime.datetime.now().strftime("%d.%b %Y %H:%M:%S")
print(detector.GetPredictionLabel(prediction) + " " + now)
os.system(play_command + " ./resources/ding.wav")
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)
def label_stream(labels, libpath, graph, sensitivity):
audio_stream = AudiostreamSource()
detector = AudioRecognition(libpath, graph, labels)
detector.SetSensitivity(sensitivity)
bufsize = detector.GetInputDataSize()
play_command = "play -q" if platform.system() == "Darwin" else "aplay"
print("Audio Recognition Version: " + detector.GetVersionString())
audio_stream.start()
try:
while (True):
frame = audio_stream.read(bufsize, bufsize)
if (not frame):
time.sleep(0.01)
continue
prediction = detector.RunDetection(frame)
if (prediction):
now = datetime.datetime.now().strftime("%d.%b %Y %H:%M:%S")
print(detector.GetPredictionLabel(prediction) + " " + now)
os.system(play_command + " ./ding.wav")
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)
def label_stream(labels, libpath, verification_path, graph, sensitivity):
last_frames = []
#Keyword spotting has 200ms frames, Verifiyer takes 2 seconds of audio
max_last_frames = 10
audio_stream = AudiostreamSource()
extractor = FeatureExtractor(libpath)
detector = AudioRecognition(libpath, graph, labels)
detector.SetSensitivity(sensitivity)
verifiyer = SpeakerVerification(libpath, verification_path)
bufsize = detector.GetInputDataSize()
print("Bufsize: " + str(bufsize))
print("Audio Recognition Version: " + detector.GetVersionString())
print(
"WARNING EXPERIMENTAL: The voice verification module can be use to verify if"
)
print(
"A command is issued by a certian speaker. It processes speech signals with a"
)
print("two second length. This experimental version isn't very good yet.")
print(
"\n\n During enrolling a fingerprint of your voice is caputred. By default 5 samples"
)
print(
"Will be captured and averaged. The progam will output a similarity score between 0 and 1"
)
print("A value of 1 means totally similar, 0 means different.")
print("Currently a threshold of 0.95 seems good")
print(
"This module should not be run on a Pi Zero, as it uses excessive CPU")
print(
"Verification can also be helpful to reduce false positives of non speech signals"
)
audio_stream.start()
try:
while (True):
frame = audio_stream.read(bufsize * 2, bufsize * 2)
if (not frame):
time.sleep(0.01)
continue
features = extractor.signal_to_mel(frame)
last_frames.append(features)
if len(last_frames) > max_last_frames:
last_frames.pop(0)
prediction = detector.RunDetection(features)
if (prediction):
now = datetime.datetime.now().strftime("%d.%b %Y %H:%M:%S")
print(detector.GetPredictionLabel(prediction) + " " + now)
os.system(play_command + " ./resources/ding.wav")
detect_frame = bytearray()
for element in last_frames:
detect_frame.extend(element)
print("Running Verification")
features = verifiyer.VerifySpeaker(detect_frame)
if (len(fingerprints) < enrolling):
print("Enrolling")
fingerprints.append(features)
else:
print("Completed")
print(features)
avg_fingerprint = get_averaged_fingerprint()
if (avg_fingerprint):
similarity_score = cosine_similarity(
features, avg_fingerprint)
print("Similarity: " + str(similarity_score))
print("Verification Done")
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)