def extractEmotionsFromAudioFile(frames, params):
emotions = {}
print("Reading sound file...") # Test
waveFile = wave.open("audio.wav", 'wb')
waveFile.setnchannels(int(params['channels']))
waveFile.setsampwidth(params['sampwidth'])
waveFile.setframerate(params['rate'])
waveFile.writeframes(b''.join(frames))
waveFile.close()
(sampleRate, samples) = scipy.io.wavfile.read("audio.wav")
bufferLen = len(samples)
cBuffer = Vokaturi.SampleArrayC(bufferLen)
if samples.ndim == 1:
cBuffer[:] = samples[:] / 32768.0 # mono
else:
cBuffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
voice = Vokaturi.Voice(sampleRate, bufferLen)
voice.fill(bufferLen, cBuffer)
print("Extracting emotions from VokaturiVoice...") # Test
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
emotions["neutral"] = emotionProbabilities.neutrality
emotions["happiness"] = emotionProbabilities.happiness
emotions["sadness"] = emotionProbabilities.sadness
emotions["angry"] = emotionProbabilities.anger
emotions["fear"] = emotionProbabilities.fear
voice.destroy()
return emotions
def get_sentiment(sample_rate, samples):
print('Sample rate %.3f Hz' % sample_rate)
print('Allocating Vokaturi sample array...')
buffer_length = len(samples)
print('%d samples, %d channels' % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5*(samples[:,0]+0.0+samples[:,1]) / 32768.0 # stereo
print('Creating VokaturiVoice...')
try:
voice = Vokaturi.Voice (sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
sentiments = {
'neutral': emotionProbabilities.neutrality,
'happy': emotionProbabilities.happiness,
'sad': emotionProbabilities.sadness,
'angry': emotionProbabilities.anger,
'fearful': emotionProbabilities.fear,
}
print('Sentiments: {}'.format(sentiments))
return sentiments
finally:
voice.destroy()
def GetVoiceEmo():
# get fresh samples
data = stream.read(CHUNKSIZE)
samples = np.fromstring(data, dtype=np.int16)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5*(samples[:,0]+0.0+samples[:,1]) / 32768.0
# initialise voice with sample rate and size
voice = Vokaturi.Voice(sample_rate, buffer_length)
# assign current recorded voice into voice var
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emoProb = Vokaturi.EmotionProbabilities()
# get probabilities
voice.extract(quality, emoProb)
# destroy voice before return value
voice.destroy()
if quality.valid:
return [emoProb.neutrality, emoProb.happiness, emoProb.sadness, emoProb.anger, emoProb.fear]
print ("Neutral: %.3f" % emoProb.neutrality)
print ("Happy: %.3f" % emoProb.happiness)
print ("Sad: %.3f" % emoProb.sadness)
print("Angry: %.3f" % emoProb.anger)
print ("Fear: %.3f" % emoProb.fear)
else:
return [0,0,0,0,0]
print("Not enough sonorancy to determine emotions")
print()
def getProbabilities(blob_name):
local_file = "/tmp/" + blob_name
wav_file = "/tmp/" + blob_name + ".wav"
_blob = bucket.blob(blob_name)
_blob.download_to_filename(local_file)
subprocess.call(['ffmpeg', '-i', local_file, wav_file])
(sample_rate, samples) = wav_read(wav_file)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
data = {}
if quality.valid:
data["neutral"] = emotionProbabilities.neutrality
data["happiness"] = emotionProbabilities.happiness
data["sadness"] = emotionProbabilities.sadness
data["anger"] = emotionProbabilities.anger
data["fear"] = emotionProbabilities.fear
else:
data["error"] = "Quality Too Low"
voice.destroy()
subprocess.Popen(['rm', local_file, wav_file])
return data
def analyzeEmotion(self, filePath):
try:
(sample_rate, samples) = wavfile.read(filePath)
print(" sample rate %.3f Hz" % sample_rate)
#print (" samples %.3f Hz" % len(samples))
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
result = filePath
if quality.valid:
#print ("Neutral: %.3f" % emotionProbabilities.neutrality)
#print ("Happy: %.3f" % emotionProbabilities.happiness)
#print ("Sad: %.3f" % emotionProbabilities.sadness)
#print ("Angry: %.3f" % emotionProbabilities.anger)
#print ("Fear: %.3f" % emotionProbabilities.fear)
result = result + ";%.3f" % emotionProbabilities.neutrality + ";%.3f" % emotionProbabilities.happiness + ";%.3f" % emotionProbabilities.sadness + ";%.3f" % emotionProbabilities.anger + ";%.3f" % emotionProbabilities.fear
else:
print("Not enough sonorancy to determine emotions")
voice.destroy()
return result
except:
print(filePath)
def vokaturi_analyze(buf, sockname, cur_time):
sample_rate = 44100
samples = np.frombuffer(buf, dtype=np.int16)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
print "===================================================="
print cur_time, "Vokaturi results from " + sockname
print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)
print "===================================================="
voice.destroy()
def analyzeFile(filepath, lineNum):
print("in analyzeFile 1")
Vokaturi.load("./lib/Vokaturi_mac.so")
(sample_rate, samples) = scipy.io.wavfile.read(filepath)
print("in analyzeFile 2")
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
print("in analyzeFile 3")
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
print("in analyzeFile 4")
if quality.valid:
emotions = [
emotionProbabilities.happiness, emotionProbabilities.sadness,
emotionProbabilities.anger, emotionProbabilities.fear,
emotionProbabilities.neutrality
]
if max(emotions) == emotionProbabilities.happiness:
maxEmotion = "Happy"
elif max(emotions) == emotionProbabilities.sadness:
maxEmotion = "Sad"
elif max(emotions) == emotionProbabilities.anger:
maxEmotion = "Angry"
elif max(emotions) == emotionProbabilities.neutrality:
maxEmotion = "Neut"
else:
maxEmotion = "Afraid"
stats = (
"Happy: %.3f\tSad: %.3f\tAngry %.3f\tFear %.3f\tNeut %.3f" %
(emotions[0], emotions[1], emotions[2], emotions[3], emotions[4]))
print("in analyzeFile 5")
emotionFile = open("emotions", 'a')
print("in analyzeFile 6")
writeEmotions(emotionFile, maxEmotion + " " + stats, lineNum)
print("in analyzeFile 7")
emotionFile.close()
print("in analyzeFile 8")
else:
print("Not enough sonorancy to determine emotions")
def voice_analysis():
print("Loading library...")
Vokaturi.load("../SIM/libs/OpenVokaturi-2-1/lib/Vokaturi_mac.so")
print("Analyzed by: %s" % Vokaturi.versionAndLicense())
print("Reading sound file...")
file_name = "demo.wav"
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
print(" sample rate %.3f Hz" % sample_rate)
print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
with Connection('amqp://*****:*****@localhost:5672//') as conn:
producer = conn.Producer(serializer='json')
producer.publish(
{
'Neutral': format(emotionProbabilities.neutrality, '.3f'),
'Happy': format(emotionProbabilities.happiness, '.3f'),
'Sad': format(emotionProbabilities.sadness, '.3f'),
'Angry': format(emotionProbabilities.anger, '.3f'),
'Fear': format(emotionProbabilities.fear, '.3f')
},
exchange=media_exchange,
routing_key='voice',
declare=[voice_queue])
# print ("Neutral: %.3f" % emotionProbabilities.neutrality)
# print ("Happy: %.3f" % emotionProbabilities.happiness)
# print ("Sad: %.3f" % emotionProbabilities.sadness)
# print ("Angry: %.3f" % emotionProbabilities.anger)
# print ("Fear: %.3f" % emotionProbabilities.fear)
else:
print("Not enough sonorancy to determine emotions")
voice.destroy()
def emotion_recognition(speech):
(sample_rate, samples) = scipy.io.wavfile.read(speech)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
#extracting emotions from speech
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)
else:
print("Can't determine emotions")
emotions = {
'Neutral': emotionProbabilities.neutrality,
'Happy': emotionProbabilities.happiness,
'Sad': emotionProbabilities.sadness,
'Angry': emotionProbabilities.anger,
'Fear': emotionProbabilities.fear
}
#print(emotions)
import operator
emotion = max(emotions.items(), key=operator.itemgetter(1))[0]
value = max(emotions.items(), key=operator.itemgetter(1))[1]
print("\n\nEmotion:\n %s with probability %.3f" % (emotion, value))
#
# for i in emotions:
# print(emotionProbabilities.i)
voice.destroy()
def vokatori_fun(file_name):
#print("Loading library...")
Vokaturi.load(
"C:/Users/gokhalea/HackathonCodes/OpenVokaturi-3-0a/OpenVokaturi-3-0a/lib/open/win/OpenVokaturi-3-0-win64.dll"
)
#print("Analyzed by: %s" % Vokaturi.versionAndLicense())
#print("Reading sound file...")
#file_name = sys.argv[1]
#(sample_rate, samples) = scipy.io.wavfile.read(file_name)
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
#print(" sample rate %.3f Hz" % sample_rate)
#print("Samples:" % samples)
#print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
#print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
#print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
#print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)
print(
"______________________________________________________________________________________________"
)
return emotionProbabilities
else:
print("Not enough sonorancy to determine emotions")
voice.destroy()
return None
def getEmotionFromWav(filePath):
emotionDict = dict()
if not (os.path.exists(filePath)):
return
#reading sound file
(sample_rate, samples) = scipy.io.wavfile.read(filePath)
print(" sample rate %.3f Hz" % sample_rate)
#print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
#print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
#print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
#print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
#print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)
emotionDict["neutrality"] = float("{:.3f}".format(
emotionProbabilities.neutrality))
emotionDict["happiness"] = float("{:.3f}".format(
emotionProbabilities.happiness))
emotionDict["sadness"] = float("{:.3f}".format(
emotionProbabilities.sadness))
emotionDict["anger"] = float("{:.3f}".format(
emotionProbabilities.anger))
emotionDict["fear"] = float("{:.3f}".format(emotionProbabilities.fear))
voice.destroy()
return emotionDict
def callVokaturi(fileName):
# Loading Vokaturi Mac
Vokaturi.load("/Users/nchao/Desktop/Yale Hacks/lib/Vokaturi_mac.so")
# Reading sound files (.wav)
file_name = "/Users/nchao/Desktop/Yale Hacks/" + fileName
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
# Allocating Vokaturi sample array
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
# Creating VokaturiVoice and filling it with voice sample
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
# Extracting emotions from Vokaturi
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
emoDict = {
"Neutral": emotionProbabilities.neutrality,
"Happy": emotionProbabilities.happiness,
"Sad": emotionProbabilities.sadness,
"Angry": emotionProbabilities.anger,
"Fear": emotionProbabilities.fear
}
# Finding main emotion in voice file
sortedVals = sorted(emoDict.values())[::-1]
stdDev = numpy.std(sortedVals)
emotions = []
for percentage in sortedVals:
if percentage > abs(max(sortedVals) - 1.5 * stdDev):
emotions += [
key for key, val in emoDict.items() if val == percentage
]
voice.destroy()
return emoDict
def analyze(file_name):
# print ("Loading library...")
Vokaturi.load("../lib/Vokaturi_linux64.so")
# print ("Analyzed by: %s" % Vokaturi.versionAndLicense())
# print ("Reading sound file...")
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
# print (" sample rate %.3f Hz" % sample_rate)
# print ("Allocating Vokaturi sample array...")
buffer_length = len(samples)
# print (" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
# print ("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
# print ("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
# print ("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
enabled = True
if quality.valid:
# print ("Neutral: %.3f" % emotionProbabilities.neutrality)
# print ("Happy: %.3f" % emotionProbabilities.happiness)
# print ("Sad: %.3f" % emotionProbabilities.sadness)
# print ("Angry: %.3f" % emotionProbabilities.anger)
# print ("Fear: %.3f" % emotionProbabilities.fear)
value = emotionProbabilities.anger
else:
value = 0
enabled = False
voice.destroy()
return enabled, value
def sentiment_analysis(path=sys.argv[1]):
print("Loading library...")
Vokaturi.load("./OpenVokaturi-3-0-linux64.so")
print("Analyzed by: %s" % Vokaturi.versionAndLicense())
print("Reading sound file...")
file_name = path
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
print(" sample rate %.3f Hz" % sample_rate)
print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)
out_dict = {}
out_dict['neutral'] = emotionProbabilities.neutrality
out_dict['happy'] = emotionProbabilities.happiness
out_dict['sad'] = emotionProbabilities.sadness
out_dict['angry'] = emotionProbabilities.anger
out_dict['fear'] = emotionProbabilities.fear
voice.destroy()
return out_dict
def extract_emotions(file_path):
Vokaturi.load(get_vokaturi_lib())
(sample_rate, samples) = scipy.io.wavfile.read(file_path)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
# if quality.valid:
print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)
emotions = {
'neutrality': "%.3f" % emotionProbabilities.neutrality,
'happiness': "%.3f" % emotionProbabilities.happiness,
'sadness': "%.3f" % emotionProbabilities.sadness,
'anger': "%.3f" % emotionProbabilities.anger,
'fear': "%.3f" % emotionProbabilities.fear
}
voice.destroy()
return emotions
def vokalculate(self, soundArr, samplerate):
'''
Calculates the emotionality of a sound sample
:param double [] soundArr: Array containing the sound
:param int samplerate: samplerate of the sound to process
:return dictionary containing each emotions probability from [0,1]
, a string with a log and a boolean (true if calculations worked)
'''
error = "Starting"
buffer_size = len(soundArr)
c_buffer = Vokaturi.SampleArrayC(buffer_size)
c_buffer[:] = soundArr[:]
self.voice = Vokaturi.Voice(samplerate, buffer_size)
self.voice.fill(buffer_size, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
self.voice.extract(quality, emotionProbabilities)
success = bool(quality.valid)
if (quality.valid):
error = error + "\n SUCCESS!"
else:
error = error + "\n Not enough sonorancy to determine emotions"
return {
"Neutral": emotionProbabilities.neutrality,
"Happy": emotionProbabilities.happiness,
"Sad": emotionProbabilities.sadness,
"Angry": emotionProbabilities.anger,
"Fear": emotionProbabilities.fear,
"Error": error,
"Success": success
}
def analyze(file):
"""Computes EmotionProbabilities from the provided wave file"""
global decibel
(sample_rate, samples) = scipy.io.wavfile.read(file)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
ep = Vokaturi.EmotionProbabilities()
voice.extract(quality, ep)
a = ep if quality.valid and MAX_LOUDNESS + decibel > params.MIN_LOUDNESS else Vokaturi.EmotionProbabilities(
0, 0, 0, 0, 0)
k, b = mavg(a)
show(k, b)
set_color(get_color(b))
voice.destroy()
def analyzeAudio(file):
(sample_rate, samples) = scipy.io.wavfile.read(file)
print(" sample rate %.3f Hz" % sample_rate)
print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
myjson.append([{
"Neutral": emotionProbabilities.neutrality
}, {
"Happy": emotionProbabilities.happiness
}, {
"Sad": emotionProbabilities.sadness
}, {
"Angry": emotionProbabilities.anger
}, {
"Fear": emotionProbabilities.fear
}])
else:
print("Not enough sonorancy to determine emotions")
voice.destroy()
def analyzeAudio(filename):
print("Loading library...")
Vokaturi.load("lib/OpenVokaturi-3-3/lib/open/win/OpenVokaturi-3-3-win64.dll")
print("Analyzed by: %s" % Vokaturi.versionAndLicense())
print("Reading sound file...")
(sample_rate, samples) = scipy.io.wavfile.read(filename)
print(" sample rate %.3f Hz" % sample_rate)
print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5*(samples[:,0]+0.0+samples[:,1]) / 32768.0 # stereo
print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)
voice.destroy()
return emotionProbabilities
def analyze_emotions(file_name):
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
print(" sample rate %.3f Hz" % sample_rate)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
return (float(emotionProbabilities.neutrality),
float(emotionProbabilities.happiness),
float(emotionProbabilities.sadness),
float(emotionProbabilities.anger),
float(emotionProbabilities.fear))
else:
return (float(0), float(0), float(0), float(0), float(0))
def emotion_recognition(speech):
#import library Vokaturi
sys.path.append("../api")
Vokaturi.load("../lib/Vokaturi_mac.so")
(sample_rate, samples) = scipy.io.wavfile.read(speech)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
#extracting emotions from speech
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
'''print("Neutral: %.3f" % emotionProbabilities.neutrality)
print("Happy: %.3f" % emotionProbabilities.happiness)
print("Sad: %.3f" % emotionProbabilities.sadness)
print("Angry: %.3f" % emotionProbabilities.anger)
print("Fear: %.3f" % emotionProbabilities.fear)'''
emotion = emotionProbabilities.max
percent = emotionProbabilities.max.percent
voice.destroy()
return emotion, percent
def detectEmotion(self):
sample_rate = VocalEmotionEstimatorEngine.RATE
samples = np.fromstring(''.join(self.audio2send), dtype=np.int16)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
self.emotion = [
emotionProbabilities.neutrality,
emotionProbabilities.happiness, emotionProbabilities.sadness,
emotionProbabilities.anger, emotionProbabilities.fear
]
print self.emotion
voice.destroy()
def listen_me():
global text, duration
parser = argparse.ArgumentParser()
parser.add_argument('--filename', '-f', default='recording.wav')
args = parser.parse_args()
# 라이브러리 준비
Vokaturi.load("/home/pi/lib/piZero.so")
# 클라우드 스피치, 텍스트 자연어처리, tts 클라이언트 각각 초기화
client = CloudSpeechClient()
nlp_client = language.LanguageServiceClient()
tts_client = texttospeech.TextToSpeechClient()
pos_wavs = []
neut_wavs = []
neg_wavs = []
intro_wavs = []
pos_wavs.append(text_to_audio(tts_client, '진짜?', '0.wav'))
pos_wavs.append(text_to_audio(tts_client, '대박', '1.wav'))
pos_wavs.append(text_to_audio(tts_client, '우와', '2.wav'))
pos_wavs.append(text_to_audio(tts_client, '하하', '3.wav'))
neut_wavs.append(text_to_audio(tts_client, '응', '10.wav'))
neut_wavs.append(text_to_audio(tts_client, '그렇구나', '11.wav'))
neut_wavs.append(text_to_audio(tts_client, '그래서?', '12.wav'))
neut_wavs.append(text_to_audio(tts_client, '응응', '13.wav'))
neg_wavs.append(text_to_audio(tts_client, '저런', '4.wav'))
neg_wavs.append(text_to_audio(tts_client, '힘내', '5.wav'))
neg_wavs.append(text_to_audio(tts_client, '에휴', '6.wav'))
intro_wavs.append(text_to_audio(tts_client, '들어줄게. 얘기해봐', 'intro0.wav'))
intro_wavs.append(text_to_audio(tts_client, '무슨 일 이야?', 'intro1.wav'))
play_wav(random.choice(intro_wavs))
logging.basicConfig(level=logging.INFO)
with Board() as board:
while True:
print('말해보자.')
text = None
duration = 0.
emotion = None
def wait():
global text, duration
start = time.monotonic()
while text is None:
# 텍스트로 인식
text = client.recognize(language_code='ko-KR')
duration = time.monotonic() - start
# 녹음하면서
record_file(AudioFormat.CD,
filename=args.filename,
wait=wait,
filetype='wav')
print(text)
print('Recorded: %.02f seconds' % duration)
if text in ['들어줘서 고마워', '내 얘기 들어줘서 고마워', '어시스턴트', '잘가', '잘 가']:
return
# 텍스트 감정 분석
document = types.Document(content=text,
type=enums.Document.Type.PLAIN_TEXT)
sentiment = nlp_client.analyze_sentiment(
document=document).document_sentiment
print('텍스트 감정 분석*********************************')
print('Text: {}'.format(text))
print('Sentiment: {}, {}'.format(sentiment.score,
sentiment.magnitude))
##################### 실험후 바꿔도 됨 ####################
pos_standard = 0.6
neg_standard = 0.1
# magnitude_standard = 0.1
# text sentiment analysis is enough
if (sentiment.score < neg_standard
or sentiment.score > pos_standard):
if sentiment.score < neg_standard:
emotion = False
print("@@@negative")
else:
emotion = True
print("@@@positive")
else:
# 녹음 파일 감정 분석
print('오디오 감정 분석*********************************')
(sample_rate, samples) = scipy.io.wavfile.read(args.filename)
# print (" sample rate %.3f Hz" % sample_rate)
# print ("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" %
(buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0
# print ("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
# print ("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
# print ("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
# print ("Neutral: %.3f" % emotionProbabilities.neutrality)
# print ("Happy: %.3f" % emotionProbabilities.happiness)
# print ("Sad: %.3f" % emotionProbabilities.sadness)
# print ("Angry: %.3f" % emotionProbabilities.anger)
# print ("Fear: %.3f" % emotionProbabilities.fear)
# fear 는 무시하도록 하자.
wave_score = emotionProbabilities.happiness - (
emotionProbabilities.sadness +
emotionProbabilities.anger)
if wave_score > 0 and sentiment.score > 0.4:
print('@@@긍정')
emotion = True
elif wave_score < 0 and sentiment.score < 0.4:
print('@@@부정')
emotion = False
# text 스코어와 wave 스코어가 불일치 할때는 중립반응 (emotion = None)
# 여기서 부터 반응.
with Leds() as leds:
if emotion is True:
play_wav(random.choice(pos_wavs))
leds.pattern = Pattern.blink(100)
color = (255, 255, 0)
leds.update(Leds.rgb_pattern(color))
time.sleep(1)
# play_wav('laugh.wav')
elif emotion is False:
play_wav(random.choice(neg_wavs))
leds.pattern = Pattern.breathe(1000)
color = (102, 140, 255)
leds.update(Leds.rgb_on(color))
time.sleep(1)
# play_wav('people-cheering.wav')
# 중립 리액션
else:
play_wav(random.choice(neut_wavs))
leds.pattern = Pattern.blink(5)
color = (230, 0, 115)
leds.update(Leds.rgb_on(color))
time.sleep(1)
"\n please input the name of your next file, type 'done' when finished: "
)
files.append(file_name)
fileCount += 1
#calculates emotion probabilites from files[] - the last entry
#
for i in range(fileCount - 1):
print("Calculating...")
print('-- ' + files[i])
file_name = files[i]
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
soundQuality = Vokaturi.Quality()
emoProbs = Vokaturi.EmotionProbabilities()
voice.extract(soundQuality, emoProbs)
if soundQuality.valid:
emoNeut += emoProbs.neutrality
def processRequest(req):
#Obtain info from the query in Dialogflow
result = req.get("queryResult")
parameters = result.get("parameters")
folder_name = parameters.get("FolderType")
#verify credentials to use google drive API & get Google API client (or something like that)
service = authentication()
#verify credentials to use google sheet API & get Google sheet
wks = open_gsheet()
#check for file in drive
(file_name , file_id) = get_wav_file(folder_name,service)
if not file_name: #If None, this will be false -> then flipped to true
return {
"fulfillmentText": "No such file in drive"
}
request = service.files().get_media(fileId=file_id)
#downloads binary data of wav file and stored in a buffered stream
fh = io.BytesIO()
downloader = MediaIoBaseDownload(fh, request)
done = False
while done is False:
status, done = downloader.next_chunk()
#parse buffered stream into Vokaturi (TO-DO: FIX BYTE CONVERSION)
buffer_length = fh.getbuffer().nbytes
c_buffer = Vokaturi.SampleArrayC(buffer_length)
c_buffer[:] = fh.getvalue()
voice = Vokaturi.Voice (8000, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
#Output data from Vokaturi & save to Google sheet
output = "The results of the analysis of " + file_name + " is... "
if quality.valid:
output += 'Neutral: %.5f, ' % emotionProbabilities.neutrality
output += 'Happiness: %.5f, ' % emotionProbabilities.happiness
output += 'Sadness: %.5f, ' % emotionProbabilities.sadness
output += 'Anger: %.5f, ' % emotionProbabilities.anger
output += 'Fear: %.5f' % emotionProbabilities.fear
if wks.cell(row, 1).value == "":
wks.update_cell(row, 1, file_name)
wks.update_cell(row, 2, '%0.5f' % emotionProbabilities.neutrality)
wks.update_cell(row, 3, '%0.5f' % emotionProbabilities.happiness)
wks.update_cell(row, 4, '%0.5f' % emotionProbabilities.sadness)
wks.update_cell(row, 5, '%0.5f' % emotionProbabilities.anger)
wks.update_cell(row, 6, '%0.5f' % emotionProbabilities.fear)
else:
while wks.cell(row, 1).value != "":
row += 1
wks.update_cell(row, 1, file_name)
wks.update_cell(row, 2, '%0.5f' % emotionProbabilities.neutrality)
wks.update_cell(row, 3, '%0.5f' % emotionProbabilities.happiness)
wks.update_cell(row, 4, '%0.5f' % emotionProbabilities.sadness)
wks.update_cell(row, 5, '%0.5f' % emotionProbabilities.anger)
wks.update_cell(row, 6, '%0.5f' % emotionProbabilities.fear)
else:
output += "Not enough sonorancy to determine emotions"
voice.destroy()
return {
"fulfillmentText": output
}
def listen_me():
global text, duration
parser = argparse.ArgumentParser()
parser.add_argument('--filename', '-f', default='recording.wav')
args = parser.parse_args()
# 라이브러리 준비
Vokaturi.load("/home/pi/lib/piZero.so")
# 클라우드 스피치랑 텍스트 자연어처리 클라이언트 각각 초기화
client = CloudSpeechClient()
nlp_client = language.LanguageServiceClient()
logging.basicConfig(level=logging.INFO)
with Board() as board:
while True:
print('말해보자.')
text = None
duration = 0.
emotion = None
def wait():
global text, duration
start = time.monotonic()
while text is None:
# 텍스트로 인식
text = client.recognize(language_code='ko-KR')
duration = time.monotonic() - start
# 녹음하면서
record_file(AudioFormat.CD,
filename=args.filename,
wait=wait,
filetype='wav')
print(text)
print('Recorded: %.02f seconds' % duration)
if text in ['들어줘서 고마워', '내 얘기 들어줘서 고마워', '어시스턴트', '잘가', '잘 가']:
return
# 텍스트 감정 분석
document = types.Document(content=text,
type=enums.Document.Type.PLAIN_TEXT)
sentiment = nlp_client.analyze_sentiment(
document=document).document_sentiment
print('텍스트 감정 분석*********************************')
print('Text: {}'.format(text))
print('Sentiment: {}, {}'.format(sentiment.score,
sentiment.magnitude))
##################### 실험후 바꿔도 됨 ####################
pos_standard = 0.6
neg_standard = 0.1
# magnitude_standard = 0.1
# text sentiment analysis is enough
if (sentiment.score < neg_standard
or sentiment.score > pos_standard):
if sentiment.score < neg_standard:
emotion = False
print("@@@negative")
else:
emotion = True
print("@@@positive")
else:
# 녹음 파일 감정 분석
print('오디오 감정 분석*********************************')
(sample_rate, samples) = scipy.io.wavfile.read(args.filename)
# print (" sample rate %.3f Hz" % sample_rate)
# print ("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" %
(buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0
# print ("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
# print ("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
# print ("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
# print ("Neutral: %.3f" % emotionProbabilities.neutrality)
# print ("Happy: %.3f" % emotionProbabilities.happiness)
# print ("Sad: %.3f" % emotionProbabilities.sadness)
# print ("Angry: %.3f" % emotionProbabilities.anger)
# print ("Fear: %.3f" % emotionProbabilities.fear)
# fear 는 무시하도록 하자.
wave_score = emotionProbabilities.happiness - (
emotionProbabilities.sadness +
emotionProbabilities.anger)
if wave_score > 0:
print('@@@긍정')
emotion = True
else:
print('@@@부정')
emotion = False
# text 분석 모호하고 wave 분석 실패했을때 (주로 목소리 짧아서)
if emotion is None:
print('please say again')
# 아님 중립적 반응 넣어도 됨.
continue
# 여기서 부터 반응.
with Leds() as leds:
if emotion is True:
# tts.say('I am glad to hear that.')
# tts.say('진짜? 대박.')
leds.pattern = Pattern.blink(100)
color = (255, 255, 0)
leds.update(Leds.rgb_pattern(color))
time.sleep(1)
# play_wav('laugh.wav')
else:
# tts.say('I am sorry to hear that.')
# tts.say('저런. 힘내.')
leds.pattern = Pattern.breathe(1000)
color = (102, 140, 255)
leds.update(Leds.rgb_on(color))
time.sleep(1)
def displayProfile(self):
self.NameEntry.focus()
self.NameEntry.delete(0, END)
#self.NameEntry.insert(0, "Hi Mitali")
self.WPMEntry.focus()
self.WPMEntry.delete(0, END)
self.Audio_Length_Secs = 150 #librosa.get_duration(filename='microphone-results.wav')
self.Audio_Length_Mins = self.Audio_Length_Secs/60
self.WPM.set(len(self.wordsList)/self.Audio_Length_Mins)
self.WPMEntry.insert(0, "You are able to speak %.2f words per min" % self.WPM.get())
self.AccuracyEntry.focus()
self.AccuracyEntry.delete(0, END)
#WER = wer(self.sample_text , self.message, standardize=True)
WER, displayList = self.werCustom(self.sample_text , self.message)
formattedText = " ".join(displayList)
displayText = formattedText.replace('#!<sb>!#', '').replace('#!<se>!#', '').replace('#!<db>!#', '').replace('#!<de>!#', '').replace('#!<ib>!#', '').replace('#!<ie>!#', '').replace('#!<sbe>!#', '')
self.text.insert(INSERT, displayText)
#self.text.insert(INSERT, formattedText)
sbindexes =[m.start() for m in re.finditer('#!<sb>!#', formattedText)]
seindexes =[m.start() for m in re.finditer('#!<se>!#', formattedText)]
dbindexes =[m.start() for m in re.finditer('#!<db>!#', formattedText)]
deindexes =[m.start() for m in re.finditer('#!<de>!#', formattedText)]
ibindexes =[m.start() for m in re.finditer('#!<ib>!#', formattedText)]
ieindexes =[m.start() for m in re.finditer('#!<ie>!#', formattedText)]
sub = ""
Del = ""
Ins = ""
for i in range(len(sbindexes)):
sbText = re.findall(r'#!<sb>!#', formattedText[0:sbindexes[i]])
dbText = re.findall(r'#!<db>!#', formattedText[0:sbindexes[i]])
ibText = re.findall(r'#!<ib>!#', formattedText[0:sbindexes[i]])
sbCount = len(sbText)+len(dbText)+len(ibText)
b = sbindexes[i]-(sbCount * 16)
e = seindexes[i]-((sbCount * 16)+8)
t = displayText[seindexes[i]-((sbCount * 16)+8): ]
tagWord = displayText[sbindexes[i]-(sbCount * 16) : seindexes[i]-((sbCount * 16)+8)]
endIndex = self.text.search(t, INSERT)
startIndex = endIndex + ('-%dc' % len(tagWord))
sub = sub + "," + tagWord +"_" +str(e)+"-"+ startIndex + ":" + endIndex
self.text.tag_add("substitutes", startIndex, endIndex) # tag and select found string
self.text.focus() # select text widget itself
#self.lblSub['text'] = sub
for i in range(len(dbindexes)):
sbText = re.findall(r'#!<sb>!#', formattedText[0:dbindexes[i]])
dbText = re.findall(r'#!<db>!#', formattedText[0:dbindexes[i]])
ibText = re.findall(r'#!<ib>!#', formattedText[0:dbindexes[i]])
sbCount = len(sbText)+len(dbText)+len(ibText)
b = dbindexes[i]-(sbCount * 16)
e = deindexes[i]-((sbCount * 16)+8)
t = displayText[deindexes[i]-((sbCount * 16)+8): ]
tagWord = displayText[dbindexes[i]-(sbCount * 16) : deindexes[i]-((sbCount * 16)+8)]
endIndex = self.text.search(t, INSERT)
startIndex = endIndex + ('-%dc' % len(tagWord))
Del = Del + "," + startIndex + ":" + endIndex
self.text.tag_add("deletions", startIndex , endIndex )
for i in range(len(ibindexes)):
sbText = re.findall(r'#!<sb>!#', formattedText[0:ibindexes[i]])
dbText = re.findall(r'#!<db>!#', formattedText[0:ibindexes[i]])
ibText = re.findall(r'#!<ib>!#', formattedText[0:ibindexes[i]])
sbCount = len(sbText)+len(dbText)+len(ibText)
b = ibindexes[i]-(sbCount * 16)
e = ieindexes[i]-((sbCount * 16)+8)
t = displayText[ieindexes[i]-((sbCount * 16)+8): ]
tagWord = displayText[ibindexes[i]-(sbCount * 16) : ieindexes[i]-((sbCount * 16)+8)]
endIndex = self.text.search(t, INSERT)
startIndex = endIndex + ('-%dc' % len(tagWord))
Ins = Ins + "," + startIndex + ":" + endIndex
self.text.tag_add("insertions", startIndex , endIndex )
#self.lblSub['text'] = sub
#self.lblIns['text'] = Ins
#self.lblDel['text'] = Del
self.text.tag_config("substitutes", background="yellow", foreground="blue")
self.text.tag_config("deletions", background="red", foreground="green")
self.text.tag_config("insertions", background="green", foreground="yellow")
Acc = (1 - WER)*100
self.AccuracyEntry.insert(0, "%f percent" % Acc)
Vokaturi.load(os.path.abspath('.')+"/OpenVokaturi-3-3/lib/OpenVokaturi-3-3-mac64.dylib")
(sample_rate, samples) = scipy.io.wavfile.read('microphone-results.wav')
buffer_length = len(samples)
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5*(samples[:,0]+0.0+samples[:,1]) / 32768.0 # stereo
#print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
#print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
#print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
#print("Neutral: %.3f" % emotionProbabilities.neutrality)
#print("Happy: %.3f" % emotionProbabilities.happiness)
#print("Sad: %.3f" % emotionProbabilities.sadness)
#print("Angry: %.3f" % emotionProbabilities.anger)
#print("Fear: %.3f" % emotionProbabilities.fear)
dictemotionProbabilities = dict()
dictemotionProbabilities['Neutral'] = emotionProbabilities.neutrality
dictemotionProbabilities['Happy'] = emotionProbabilities.happiness
dictemotionProbabilities['Sad'] = emotionProbabilities.sadness
dictemotionProbabilities['Angry'] = emotionProbabilities.anger
dictemotionProbabilities['Fear'] = emotionProbabilities.fear
self.EmoEntry.focus()
self.EmoEntry.delete(0, END)
emo = max(dictemotionProbabilities, key=dictemotionProbabilities.get)
self.EmoEntry.insert(0, emo)
else:
self.EmoEntry.focus()
self.EmoEntry.delete(0, END)
self.EmoEntry.insert(0, 'could not find emotion from the voice')
voice.destroy()
def audio_thread():
p = pyaudio.PyAudio()
for ii in range(p.get_device_count()):
print(str(ii) + " " + p.get_device_info_by_index(ii).get('name'))
form_1 = pyaudio.paInt16 # 16-bit resolution
chans = 1 # 1 channel
samp_rate = 44100 # 44.1kHz sampling rate
chunk = 132096 # number of recorded samples for buffer
record_secs = 3 # seconds to record
dev_index = 2 # device index found by p.get_device_info_by_index(ii)
i = 0
while (True):
sleep(1)
# create pyaudio stream
i = i + 1
wav_output_filename = '../sounds/test' + str(
i) + '.wav' # name of .wav file
audio = pyaudio.PyAudio()
stream = audio.open(format = form_1,rate = samp_rate,channels = chans, \
input_device_index = dev_index,input = True, \
frames_per_buffer=chunk)
#print("[AUDIO] recording")
frames = []
# loop through stream and append audio chunks to frame array
for ii in range(0, int((samp_rate / chunk) * record_secs)):
data = stream.read(chunk)
frames.append(data)
#print("[AUDIO] finished recording")
# stop the stream, close it, and terminate the pyaudio instantiation
stream.stop_stream()
stream.close()
audio.terminate()
# save the audio frames as .wav file
wavefile = wave.open(wav_output_filename, 'wb')
wavefile.setnchannels(chans)
wavefile.setsampwidth(audio.get_sample_size(form_1))
wavefile.setframerate(samp_rate)
wavefile.writeframes(b''.join(frames))
wavefile.close()
#print ("[AUDIO] Filename: " + wav_output_filename)
(sample_rate, samples) = scipy.io.wavfile.read(wav_output_filename)
#print ("[AUDIO] sample rate %.3f Hz" % sample_rate)
buffer_length = len(samples)
#print ("[AUDIO] %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1: # mono
c_buffer[:] = samples[:] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 + samples[:, 1]) / 32768.0
voice = Vokaturi.Voice(sample_rate, buffer_length)
voice.fill(buffer_length, c_buffer)
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
if quality.valid:
print("[AUDIO] Neutral: %.3f" % emotionProbabilities.neutrality)
print("[AUDIO] Happy: %.3f" % emotionProbabilities.happiness)
print("[AUDIO] Sad: %.3f" % emotionProbabilities.sadness)
print("[AUDIO] Angry: %.3f" % emotionProbabilities.anger)
print("[AUDIO] Fear: %.3f" % emotionProbabilities.fear)
audio_emotions_mutex.acquire()
audio_emotions[_NEUTRAL] = emotionProbabilities.neutrality
audio_emotions[_HAPPINESS] = emotionProbabilities.happiness
audio_emotions[_SADNESS] = emotionProbabilities.sadness
audio_emotions[_ANGER] = emotionProbabilities.anger
audio_emotions[_FEAR] = emotionProbabilities.fear
audio_emotions_mutex.release()
else:
print("[AUDIO] Not enough sonorancy to determine emotions")
voice.destroy()
def voice_predict():
file_name = "Audio/audio.wav"
print("Loading library...")
Vokaturi.load("OpenVokaturi-2-1d/lib/Vokaturi_mac64.so")
print("Analyzed by: %s" % Vokaturi.versionAndLicense())
print("Reading sound file...")
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
print(" sample rate %.3f Hz" % sample_rate)
print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(buffer_length)
if samples.ndim == 1:
c_buffer[:] = samples[:] / 32768.0 # mono
else:
c_buffer[:] = 0.5 * (samples[:, 0] + 0.0 +
samples[:, 1]) / 32768.0 # stereo
print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
print("Filling VokaturiVoice with samples...")
voice.fill(buffer_length, c_buffer)
print("Extracting emotions from VokaturiVoice...")
quality = Vokaturi.Quality()
emotionProbabilities = Vokaturi.EmotionProbabilities()
voice.extract(quality, emotionProbabilities)
fh = open("output.txt", 'a')
fh.write("Based on your voice, your emotion is ")
if not quality.valid:
fh.write("beyond my understanding.\n")
exit(1)
print_result = [(round(emotionProbabilities.neutrality*100), "neutral"), \
(round(emotionProbabilities.happiness*100), "happy"), \
(round(emotionProbabilities.anger*100), "angry"), \
(round(emotionProbabilities.fear*100),"fearful"), \
(round(emotionProbabilities.sadness*100), "sad")]
print_result = [tup for tup in print_result if tup[0] != 0]
print_result.sort(key=lambda tup: tup[0])
if len(print_result) == 0:
fh.write("beyond my understanding.\n")
elif len(print_result) == 1:
fh.write("dominantly %d percent %s.\n" %
(print_result[0][0], print_result[0][1]))
else:
for i in range(len(print_result) - 1, 0, -1):
fh.write("%d percent %s, " %
(print_result[i][0], print_result[i][1]))
fh.write("and %d percent %s.\n" %
(print_result[0][0], print_result[0][1]))
fh.close()
with open("output.txt") as f1:
with open("templates.yaml", "w") as f2:
f2.write("welcome: Ready to hear your comments?\n\nround: ")
for line in f1:
f2.write(line.strip("\n"))
f2.write(" ")
voice.destroy()
sys.path.append("../api")
import Vokaturi
print("Loading library...")
Vokaturi.load("../lib/open/macos/OpenVokaturi-3-3-mac64.dylib")
print("Analyzed by: %s" % Vokaturi.versionAndLicense())
print("Reading sound file...")
file_name = sys.argv[1]
(sample_rate, samples) = scipy.io.wavfile.read(file_name)
print(" sample rate %.3f Hz" % sample_rate)
print("Allocating Vokaturi sample array...")
buffer_length = len(samples)
print(" %d samples, %d channels" % (buffer_length, samples.ndim))
c_buffer = Vokaturi.SampleArrayC(sample_rate)
print("Creating VokaturiVoice...")
voice = Vokaturi.Voice(sample_rate, buffer_length)
numberOfSeconds = int(buffer_length / sample_rate)
print("Start(s) End(s) Neutral Happy Sad Angry Fear")
for isecond in range(0, numberOfSeconds):
startSample = round(isecond * sample_rate)
endSample = round((isecond + 1) * sample_rate)
print(startSample, endSample)
if samples.ndim == 1: # mono
c_buffer[:] = samples[startSample:endSample] / 32768.0
else: # stereo
c_buffer[:] = 0.5 * (samples[startSample:endSample, 0] + 0.0 +
samples[startSample:endSample, 1]) / 32768.0