def detectKeywords(libpath):
audio_stream = AudiostreamSource()
extractor = FeatureExtractor(libpath)
detector = AudioRecognition(libpath)
extactor_gain = 1.0
vad_threshold = 0.2
keywordVAD = detector.addContinousModel('../../models/Hotword/vad_16.premium')
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.signalToMel(frame,extactor_gain)
_ = detector.runDetection(features)
vadResult = detector.getContinousResult(keywordVAD)
if(vadResult[1] > vad_threshold):
print("Speech detected")
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)
def detectKeywords(libpath):
audio_stream = AudiostreamSource()
extractor = FeatureExtractor(libpath)
detector = AudioRecognition(libpath)
extactor_gain = 1.0
#Add one or more keyword models
keywordIdFirefox = detector.addModel(
'../../models/Hotword/firefox_v1.4.5.premium', 0.6)
keywordIdSheila = detector.addModel(
'../../models/Hotword/sheila_v1.4.5.premium', 0.6)
keywordIdMarvin = detector.addModel(
'../../models/Hotword/marvin_v1.4.5.premium', 0.6)
keywordIdAlexa = detector.addModel(
'../../models/Hotword/alexa_v1.4.5.premium', 0.6)
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.signalToMel(frame, extactor_gain)
prediction = detector.runDetection(features)
if (prediction != 0):
now = datetime.datetime.now().strftime("%d.%b %Y %H:%M:%S")
if (prediction == keywordIdFirefox):
print("Firefox detected:" + now)
elif (prediction == keywordIdSheila):
print("Sheila detected:" + now)
elif (prediction == keywordIdMarvin):
print("Marvin detected:" + now)
elif (prediction == keywordIdAlexa):
print("Alexa detected:" + now)
os.system(play_command + " ../resources/ding.wav")
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)
def recordActivations(libpath):
audio_stream = AudiostreamSource()
extractor = FeatureExtractor(libpath)
detectors = {}
framebuffersFront = {}
framebuffersBack = {}
extactor_gain = 1.0
recordBefore = 2.5 # Seconds before the activation
recordAfter = 0.5 # Seconds after the activation
activationCount = 0
ensure_dir(saveDirectory)
rbFrontSize = int(recordBefore * bytesPerSample * framesPerSecond)
rbBackSize = int(recordAfter * bytesPerSample * framesPerSecond)
for mpath, msens, mname in models:
detector = AudioRecognition(libpath)
detector.addModel(mpath, msens)
detectors[mname] = detector
framebuffersFront[mname] = bytearray()
framebuffersBack[mname] = bytearray()
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
for mname in detectors:
#Fill audio before the activation
framebuffersFront[mname] = framebuffersFront[mname] + frame
if (len(framebuffersFront[mname]) > rbFrontSize):
framebuffersFront[mname] = framebuffersFront[mname][
-rbFrontSize:]
features = extractor.signalToMel(frame, extactor_gain)
for mname in detectors:
detector = detectors[mname]
prediction = detector.runDetection(features)
if (prediction != 0):
#FIXME: Record after is currently ignored
#Fill audio after the activation
#while(len(framebuffersBack[mname]) < rbBackSize):
# frame = audio_stream.read(bufsize*2,bufsize*2)
# if(not frame):
# time.sleep(0.01)
# continue
# framebuffersBack[mname] = framebuffersBack[mname] + frame
savePath = saveDirectory + "/activation_{}_{}_{}.wav".format(
mname, activationCount, time.time_ns())
save_wav(framebuffersFront[mname], savePath)
#save_wav(framebufferFront+framebufferBack,savePath)
print("Saving Activation to {}".format(savePath))
activationCount += 1
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)
def detectKeywords(libpath):
audio_stream = AudiostreamSource()
extractor = FeatureExtractor(libpath)
detector = AudioRecognition(libpath)
framerate = 16000
model = Model("model")
#Let's define a custom dictionary
rec = KaldiRecognizer(
model, framerate,
'["oh one two three four five six seven eight nine zero", "[unk]"]')
extactor_gain = 1.0
#Add one or more keyword models
keywordIdAlexa = detector.addModel(
'../../models/Hotword/alexa_v3.0.35.premium', 0.85)
bufsize = detector.getInputDataSize()
print("Audio Recognition Version: " + detector.getVersionString())
command_started = False
audio_stream.start()
try:
while (True):
# Wakeword loop
if (not command_started):
frame = audio_stream.read(bufsize * 2, bufsize * 2)
if (not frame):
time.sleep(0.01)
continue
features = extractor.signalToMel(frame, extactor_gain)
prediction = detector.runDetection(features)
if (prediction != 0):
now = datetime.datetime.now().strftime("%d.%b %Y %H:%M:%S")
if (prediction == keywordIdAlexa):
print("Alexa detected:" + now)
os.system(play_command + " ../resources/ding.wav")
command_started = True
# vosk loop
else:
frame = audio_stream.read(4000, 4000)
if (not frame):
time.sleep(0.01)
continue
if rec.AcceptWaveform(bytes(frame)):
print(rec.Result())
command_started = False
print(rec.FinalResult())
except KeyboardInterrupt:
print("Terminating")
audio_stream.stop()
sys.exit(0)