def run(self):
#global my_mutex
global numRecordings
global audioNum
global processNum
megaArray = []
#while processNum < numRecordings:
while goingFlag == True or processNum < audioNum:
while processNum >= audioNum:
y = 0
#print processNum
if processNum > 0:
subprocess.check_output(['rm', wavFile])
#jsonFile = "/home/pi/results" + `processNum` + ".json"
wavFile = "/home/pi/talking" + ` processNum ` + ".wav"
print "Processing " + wavFile
foFile = "fo" + ` processNum ` + ".fo"
pmFile = "pm" + ` processNum ` + ".pm"
timeEnd = q.get()
print timeEnd
#print subprocess.check_output(['python', '/home/pi/detectVoiceInWav.py', wavFile, jsonFile, str(timeEnd)])
v = VoiceActivityDetector(wavFile)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(
raw_detection, str(timeEnd))
megaArray.append(speech_labels)
save_to_file(megaArray, soundName)
#print subprocess.check_output(['./reaper/REAPER/build/reaper', '-i', wavFile, '-f', foFile, '-p', pmFile, '-a'])
processNum = processNum + 1
def getSpeechStartTime(self, fileName):
v = VoiceActivityDetector(fileName)
windows = v.detect_speech()
for i in range(0, len(windows)):
arr = windows[i]
if arr[len(arr) - 1] == 1:
# VAD breaks wav file into 20 ms windowed chunks, with a 10 ms overlay
return i * 10
return -1
def get_score(file_name):
v = VoiceActivityDetector(file_name)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
start = 0
end = 0
if len(speech_labels) > 0:
start = speech_labels[0]['speech_begin']
end = speech_labels[-1]['speech_end']
return end - start
def apply_VAD(phonation_files, new_wav_path):
# speech_HD = {0:[0,0], 2:[0,0], 3:[0,0]}
# for wav in phonation_files:
# optimal_thresh = 0.5
# v = VoiceActivityDetector(wav)
# v.speech_energy_threshold = optimal_thresh
# frames = v.detect_speech()
# speech_labels = v.convert_windows_to_readible_labels(frames)
# # print('spkr wav', wav.split('/')[-1])
# # print(speech_labels)
# # exit()
# spkr_id = wav.split('/')[-1].split('_')[0]
# if len(speech_labels) > 1: #if more than 1 split
# speech_HD[early_late_balanced[spkr_id]][1]+=1
# else:
# speech_HD[early_late_balanced[spkr_id]][0]+=1
# print(speech_HD)
# exit()
for thresh_sweep in [0.25, 0.3, 0.35]: #0.5 too high, no speech detected for some spkrs
for wav in phonation_files:
optimal_thresh = thresh_sweep
v = VoiceActivityDetector(wav)
v.speech_energy_threshold = optimal_thresh
frames = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(frames)
#output new wav
old_basename = wav.split('/')[-1]
new_wav_path_dir = os.path.join(new_wav_path, str(thresh_sweep))
if not os.path.exists(new_wav_path_dir):
os.makedirs(new_wav_path_dir)
out_wav = os.path.join(new_wav_path_dir, old_basename)
# print(frames)
# for i in frames:
# if i[1] > 0:
# print(i)
# print('wav', wav)
# print(speech_labels)
# print(thresh_sweep)
if '38717' in wav:
print(speech_labels)
exit()
first_vad_instance = speech_labels[0]
beg_t = first_vad_instance['speech_begin']
end_t = first_vad_instance['speech_end']
transformer = sox.Transformer()
transformer.trim(float(beg_t),float(end_t))
transformer.convert(samplerate=32000, n_channels=1)
transformer.build(wav, out_wav)
def human_voice_detect(wave_file):
# noise reduction
denoised_wave = denoise(wave_file)
# band pass filter
perform_bandpass(denoised_wave)
# run VAD
v = VoiceActivityDetector('filtered_file.wav')
array = v.detect_speech()
#print array
return parse_voice_array(array.tolist())
def run(self):
#global my_mutex
global numRecordings
global audioNum
global processNum
megaArray = []
currDay = 1
fileChar = "A"
#while processNum < numRecordings:
while goingFlag == True or processNum < audioNum:
while processNum >= audioNum:
y=0
#print processNum
try:
if processNum > 0:
subprocess.check_output(['rm', wavFile])
#jsonFile = "/home/pi/results" + `processNum` + ".json"
wavFile = dir_path + "/talking" + `processNum` + ".wav"
print "Processing " + wavFile
#foFile = "fo" + `processNum` + ".fo"
#pmFile = "pm" + `processNum` + ".pm"
timeEnd = q.get()
tempDay = currDay
currDay = qSoundJson.get()
if tempDay != currDay:
megaArray = []
soundName = dir_path + "/data/day" + str(currDay) + "/" + fileChar + "/sound" + str(fileNum) + ".json"
if fileChar == "A":
fileChar = "B"
else:
fileChar = "A"
print timeEnd
#print subprocess.check_output(['python', '/home/pi/detectVoiceInWav.py', wavFile, jsonFile, str(timeEnd)])
v = VoiceActivityDetector(wavFile)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection, str(timeEnd))
megaArray.append(speech_labels)
save_to_file(megaArray, soundName)
#print subprocess.check_output(['./reaper/REAPER/build/reaper', '-i', wavFile, '-f', foFile, '-p', pmFile, '-a'])
processNum = processNum + 1
except subprocess.CalledProcessError as e:
logging.exception("message")
sense.clear()
with open("processWav.txt", "w") as text_file:
text_file.write("Total wavs processed: {}".format(str(processNum)))
with open("createWav.txt", "w") as t2:
t2.write("Total wavs supposed to be processed: {}".format(str(audioNum)))
def containsSpeech(inputfile, saveToFileFlag=False, outputfile='temp.json'):
#if __name__ == "__main__":
#parser = argparse.ArgumentParser(description='Analyze input wave-file and save detected speech interval to json file.')
#parser.add_argument('inputfile', metavar='INPUTWAVE',help='the full path to input wave file')
#parser.add_argument('outputfile', metavar='OUTPUTFILE',help='the full path to output json file to save detected speech intervals')
#args = parser.parse_args()
v = VoiceActivityDetector(inputfile)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
if (saveToFileFlag):
save_to_file(speech_labels, outputfile)
if (len(speech_labels) != 0):
return True
else:
return False
def detect_speech(input_file, output_file="results.txt"):
try:
v = VoiceActivityDetector(input_file)
raw_detection = v.detect_speech()
logger.info("Speech detection complete")
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
speech = get_speech_duration(speech_labels)
logger.info(
"File: {}, Duration of video: {}, Duration of speech: {}".format(
input_file, v.duration, speech))
empty_venue = (speech / v.duration) < 0.05
save_to_file(empty_venue, output_file, True)
logger.info("File: {}, Empty: {}".format(input_file, empty_venue))
except Exception as e:
save_to_file(False, output_file, False)
logger.error(
"Something went wrong while trying to detect speech: {}".format(e))
def vad_trial(file):
# Create our detector
v = VoiceActivityDetector(file)
# Generate VAD intervals for the file
speech_labels = v.convert_windows_to_readible_labels(v.detect_speech())
# Get file info
data = mediainfo(file)
# Add all the times together
total_time = 0
for speech_data in speech_labels:
# add the length of the speech
total_time += speech_data['speech_end'] - speech_data['speech_begin']
percent = total_time / float(data['duration'])
return percent
def __init__(self, wave_input_filename, output_File, duration_of_video):
v = VoiceActivityDetector(wave_input_filename)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
speech_labels = [float(i) for i in speech_labels]
if len(speech_labels) == 0:
with open(output_File, 'w') as fp:
fp.write(str(0.00) + " " + str(duration_of_video) + "\n")
return
final = []
if speech_labels[0] != 0.00:
final.append(0.00)
final.append(speech_labels[0] - 0.01)
count = 1
length = duration_of_video
last = len(speech_labels) - 1
while count <= last:
final.append(speech_labels[count] + 0.01)
count += 1
if count >= last:
final.append(length)
else:
final.append(speech_labels[count] - 0.01)
count += 1
count = 1
while count < len(final):
if final[count] - final[
count -
1] < 4: # 4 seconds is the minimum limit for caption, as sentence can't fit in less than 4 seconds
del final[count]
del final[count - 1]
else:
count += 2
if len(final) != 0:
with open(output_File, 'w') as fp:
index = 0
while index < len(final):
fp.write(
str(final[index]) + " " + str(final[index + 1]) + "\n")
index += 2
def split_wavs(inputfile, basedir):
v = VoiceActivityDetector(inputfile)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
out_file_list = []
begin = speech_labels[0]["speech_begin"]
last_end = speech_labels[0]["speech_end"]
valid_begin = begin
valid_end = last_end
num = len(speech_labels)
i = 0
for labels in speech_labels[1:]:
i += 1
end = labels["speech_end"]
if end - begin > 59:
valid_end = last_end
valid_begin = begin
print valid_begin, valid_end
outputfile = os.path.join(basedir, "out_"+str(i)+".wav")
save_wav(valid_begin, valid_end, inputfile, outputfile)
out_file_list.append(outputfile)
elif end - begin > 50:
valid_end = end
valid_begin = begin
print valid_begin, valid_end
outputfile = os.path.join(basedir, "out_"+str(i)+".wav")
save_wav(valid_begin, valid_end, inputfile, outputfile)
out_file_list.append(outputfile)
if i < num:
begin = speech_labels[i+1]["speech_begin"]
if valid_end < speech_labels[-1]["speech_end"]:
valid_begin= begin
valid_end = speech_labels[-1]["speech_end"]
print begin , speech_labels[-1]["speech_end"]
outputfile = os.path.join(basedir, "out_"+str(i)+".wav")
save_wav(valid_begin, valid_end, inputfile, outputfile)
out_file_list.append(outputfile)
return out_file_list
def main():
fn_s = glob.glob(
'./wav/171219101741/*.wav') # set the wave data directory.
now = datetime.datetime.now().strftime('%y%m%d%H%M%S')
save_path = os.path.join('./note_annotation', now)
os.makedirs(save_path)
for fn_full in fn_s:
fn, ext = os.path.splitext(fn_full)
v = VoiceActivityDetector(fn + ext)
dw, sve, pe = v.detect_speech(
) # dw: detected windows; sve: sum_voice_energy_s; pe: peak energy
st = v.convert_windows_to_readible_labels(
dw
) # return the list of speech times based on the detected window lists.
data_passed_hilbert_sum = data2hilbert(v.data,
v.rate) # hilbert transform
note_time = split_frame_hilbert_peak(dw, st, data_passed_hilbert_sum)
# here set frame length for putting in mfcc compute 0.1
df_note_time = pd.DataFrame(
np.vstack(
(note_time[0][:], note_time[0][:] + 0.1, note_time[2][:])).T,
columns=["start_time", "end_time", "phrase_label"])
df_note_time_sel = df_note_time[df_note_time.phrase_label > 0]
df_note_time_sel.reset_index()
df_note_time_sel.to_csv(save_path + '/' + fn.split('/')[-1] +
'_phrase_peak_time.csv')
mfcc_s_str_100_wo = ''
for i in range(len(df_note_time_sel.index)):
mfcc_s_100 = get_mfcc_s(v.data,
v.rate,
df_note_time_sel['start_time'].iloc[i],
df_note_time_sel['end_time'].iloc[i],
frames=0.1,
overlap=0.1)
mfcc_s_str_100_wo += mfcc2str(mfcc_s_100) + '\n'
with open(
save_path + '/' + fn.split('/')[-1] +
'_100msec_no_overlap_at_peak_loc.ghmm', 'w') as f_samples_100:
f_samples_100.write(mfcc_s_str_100_wo)
print('done for %s' % fn)
def main():
parser = argparse.ArgumentParser(
description=
'Analyze input wave-file and save detected speech interval to json file.'
)
parser.add_argument('inputfile',
metavar='INPUTWAVE',
help='the full path to input wave file')
parser.add_argument(
'outputfile',
metavar='OUTPUTFILE',
help=
'the full path to output json file to save detected speech intervals')
args = parser.parse_args()
v = VoiceActivityDetector(args.inputfile)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
print("")
print("Saving Detected Vocal Ranges to disk")
save_to_file(speech_labels, args.outputfile)
print("Done")
def main(wave_file, txt_output):
wr = wave.open(wave_file, 'r')
par = list(wr.getparams())
par[3] = 0
filtered_wav = wave_file[:-4]
filtered_wav = filtered_wav+"_fltrd.wav"
ww = wave.open(filtered_wav, 'w')
ww.setparams(tuple(par))
lowpass = 300
highpass = 3400
sz = wr.getframerate()
c = int(wr.getnframes()/sz)
for num in range(c):
print('Processing {}/{} s'.format(num+1, c))
da = np.fromstring(wr.readframes(sz), dtype=np.int16)
left, right = da[0::2], da[1::2] # left and right channel
lf, rf = np.fft.rfft(left), np.fft.rfft(right)
lf[:lowpass], rf[:lowpass] = 0, 0 # low pass filter
lf[55:66], rf[55:66] = 0, 0 # line noise
lf[highpass:], rf[highpass:] = 0,0 # high pass filter
nl, nr = np.fft.irfft(lf), np.fft.irfft(rf)
ns = np.column_stack((nl,nr)).ravel().astype(np.int16)
ww.writeframes(ns.tostring())
wr.close()
ww.close()
v = VoiceActivityDetector(filtered_wav)
with open(txt_output, "w") as open_file:
array = v.detect_speech()
open_file.write(str(array))
open_file.close()
def recognize_speech(audiofile):
from vad import VoiceActivityDetector
v = VoiceActivityDetector(audiofile)
detected = v.detect_speech()
# detected2 = v.convert_windows_to_readible_labels(detected)
return detected # returns array of window numbers and speech flags (1 - speech, 0 - nonspeech)
def main(wave_file):
v = VoiceActivityDetector(wave_file)
array = v.detect_speech()
print array.tolist()
from vad import VoiceActivityDetector
import argparse
import json
def save_to_file(data, filename):
with open(filename, 'w') as fp:
json.dump(data, fp)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Analyze input wave-file and save detected speech interval to json file.')
# parser.add_argument('inputfile', metavar='INPUTWAVE',
# help='the full path to input wave file')
parser.add_argument('outputfile', metavar='OUTPUTFILE',
help='the full path to output json file to save detected speech intervals')
args = parser.parse_args()
v = VoiceActivityDetector('rec_unlimited_7iwka57k.wav')
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
save_to_file(speech_labels, args.outputfile)
def save_to_file(data, filename):
with open(filename, 'w') as fp:
json.dump(data, fp)
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description=
'Analyze input wave-file and save detected speech interval to json file.'
)
# parser.add_argument('inputfile', metavar='INPUTWAVE',
# help='the full path to input wave file')
parser.add_argument(
'outputfile',
metavar='OUTPUTFILE',
help=
'the full path to output json file to save detected speech intervals')
args = parser.parse_args()
# Create a speech detection object and read in the audio data
v = VoiceActivityDetector('FILE_NAME.wav')
# Detect speech in the given data
raw_detection = v.detect_speech()
# Convert the detected speech signals to a readable format
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
# Save the converted and detected speech to a file
save_to_file(speech_labels, args.outputfile)
import argparse
import json
def save_to_file(data, filename):
with open(filename, 'w') as fp:
json.dump(data, fp)
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description=
'Analyze input wave-file and save detected speech interval to json file.'
)
parser.add_argument('inputfile',
metavar='INPUTWAVE',
help='the full path to input wave file')
parser.add_argument(
'outputfile',
metavar='OUTPUTFILE',
help=
'the full path to output json file to save detected speech intervals')
args = parser.parse_args()
v = VoiceActivityDetector(args.inputfile)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
print(speech_labels)
save_to_file(speech_labels, args.outputfile)
def includes_speech(file):
v = VoiceActivityDetector(file)
raw_detection = v.detect_speech()
return v.includes_speech(raw_detection, thresh=3)
EAnalysis = load_model("../Emotion_analysis")
sad = np.array([["Image", "Emotion"]], dtype=object)
predit = np.array(
[['Angry', 'Disgust', 'Fear', 'Happy', 'Sad', 'Surprise', 'Neutral']],
dtype=object)
for file in os.listdir(he.FACEDIR):
face = os.path.join(he.FACEDIR, file)
img = cv2.imread(face)
lastimg = cv2.resize(img, (48, 48))
gray_image = lastimg / 255
out = gray_image[np.newaxis, :, :]
predictions = EAnalysis.predict(out)
emotion = he.getEmotion(predictions)
sad = np.append(sad, [[face, emotion]], axis=0)
predit = np.append(predit, predictions, axis=0)
f = open('emotion.dat', 'ab')
np.savetxt(f, sad, fmt='%s')
f.close()
x = open('prediction.dat', 'ab')
np.savetxt(x, predit, fmt='%s')
x.close()
v = VoiceActivityDetector("audio.wav")
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
he.save_to_file(speech_labels, "vad.json")
def main():
datadir = './dataset/train/audio/'
labels = {
'yes': 0,
'no': 1,
'up': 2,
'down': 3,
'left': 4,
'right': 5,
'on': 6,
'off': 7,
'go': 8,
'stop': 9
}
labels_list = np.array([
'yes', 'no', 'up', 'down', 'left', 'right', 'on', 'off', 'go', 'stop'
])
train_in = []
train_out = []
test_in = []
test_out = []
removal_thr = 0
vadcount = 0
for word, label in labels.items():
print(word)
dirpath = datadir + word
files = os.listdir(dirpath)
np.random.shuffle(files)
for i, fn in enumerate(files):
filepath = os.path.join(dirpath, fn)
sample_rate, samples = wavfile.read(filepath)
#print(filepath)
# run voice activity detector
v = VoiceActivityDetector(filepath)
detected_windows = v.detect_speech()
speechtime = v.convert_windows_to_readible_labels(detected_windows)
silence_free = np.array([])
if len(speechtime) != 0:
for segment in speechtime:
n0 = int(segment['speech_begin'] * sample_rate)
n1 = int(segment['speech_end'] * sample_rate)
silence_free = np.append(silence_free, samples[n0:n1])
#print(len(silence_free))
# !!! IMPORTANT check the quality of silence removel
# bad ones will crush the spectrogram step
if len(silence_free) >= 3200:
vadcount += 1
freqs, times, spectrogram = log_specgram(silence_free,
sample_rate,
window_size=10,
step_size=5)
else:
freqs, times, spectrogram = log_specgram(samples,
sample_rate,
window_size=10,
step_size=5)
#plt.pcolormesh(freqs, times, spectrogram, cmap='coolwarm')
#plt.show()
# average spectrogram shape is (97.3,161) vvvvv: x and y are reversed in cv2
spectrogram = cv2.resize(spectrogram,
dsize=(161, 100),
interpolation=cv2.INTER_CUBIC)
spectrogram = spectrogram.reshape(100, 161, 1)
if i < len(files) / 5:
test_in.append(spectrogram)
test_out.append(one_hot_encode(label))
else:
train_in.append(spectrogram)
train_out.append(one_hot_encode(label))
train_in = np.array(train_in, dtype=np.float32)
train_out = np.array(train_out, dtype=np.int32)
test_in = np.array(test_in, dtype=np.float32)
test_out = np.array(test_out, dtype=np.int32)
print(train_in.shape)
print(train_out.shape)
print(test_in.shape)
print(test_out.shape)
print(vadcount, ' audios were processed with VAD.')
save_filename = './mini_speech_data2.npz'
np.savez(save_filename,
train_in=train_in,
train_out=train_out,
test_in=test_in,
test_out=test_out,
labels=labels_list)
from pydub import AudioSegment
from vad import VoiceActivityDetector
sound = AudioSegment.from_wav('Test1-master/Test/mic1.wav')
slicedSounds = sound[::500]
slicedSounds = list(slicedSounds)
for i in range(0, len(slicedSounds)):
slicedSounds[i].export('trash.wav', format='wav')
obj = VoiceActivityDetector('trash.wav')
data = obj.detect_speech()
Found = False
for i in range(0, len(data)):
if data[i][len(data[i]) - 1] == 1:
speech = AudioSegment.from_wav('trash.wav')
speech.export('speech.wav', format='wav')
Found = True
if Found:
break
child_puzzle_wav = FLAGS.child_puzzle_wav
mom_puzzle_wav = FLAGS.mom_puzzle_wav
mom_puzzle_textgrid = FLAGS.mom_puzzle_textgrid
child_outfile_textgrid = FLAGS.child_outfile_textgrid
add_seconds_at_boundary = FLAGS.add_seconds_at_boundary
child_segment_wav_outdir = FLAGS.child_segment_wav_outdir
mom_segment_wav_outdir = FLAGS.mom_segment_wav_outdir
if not os.path.exists(child_segment_wav_outdir):
os.makedirs(child_segment_wav_outdir)
if not os.path.exists(mom_segment_wav_outdir):
os.makedirs(mom_segment_wav_outdir)
# detects child speech parts
v = VoiceActivityDetector(child_puzzle_wav)
data = v.data
total_time = len(data) * 1.0 / v.rate
total_time = float("{0:.2f}".format(total_time))
speech_time, mom_tier = child_speech_detector(mom_puzzle_textgrid, v)
# export detected child speech segments wav
turns = export_child_audio_segments(total_time, child_puzzle_wav,
add_seconds_at_boundary,
child_segment_wav_outdir, speech_time)
total_turns = turns
tier = write_to_txtgrids('Machine-Label-CS', turns)
# modify manually annotated mom speech segments, and export the wav segments
mom_turns = export_mom_audio_segments(mom_puzzle_wav, mom_tier,
mom_segment_wav_outdir)
from vad import VoiceActivityDetector
import argparse
import json
def save_to_file(data, filename):
with open(filename, 'w') as fp:
json.dump(data, fp)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Analyze input wave-file and save detected speech interval to json file.')
parser.add_argument('inputfile', metavar='INPUTWAVE',
help='the full path to input wave file')
parser.add_argument('outputfile', metavar='OUTPUTFILE',
help='the full path to output json file to save detected speech intervals')
args = parser.parse_args()
v = VoiceActivityDetector(args.inputfile)
raw_detection = v.detect_speech()
speech_labels = v.convert_windows_to_readible_labels(raw_detection)
save_to_file(speech_labels, args.outputfile)
x = AudioSegment.from_wav(fullFileName)
x = x.set_channels(1)
x.export(FILE_FOLDER + FILE_NAME + '-mono.wav', format="wav")
waveFile = wave.open(FILE_FOLDER + FILE_NAME + '-mono.wav', 'r')
# Getting some variables from file
signal = waveFile.readframes(-1)
signal = np.fromstring(signal, 'Int16')
frameRate = waveFile.getframerate()
Time = np.linspace(0, len(signal) / frameRate, num=len(signal))
totalFrames = waveFile.getnframes()
wavFileLengthSec = totalFrames / frameRate
# VOICE ACTIVITY DETECTION (VAD)
vad = VoiceActivityDetector(fullFileName)
raw_detection = vad.detect_speech()
speech_labels, starters, enders = vad.convert_windows_to_readible_labels(
raw_detection)
# list of beginnings of voice detected segments (in frames)
starterFrames = [int(i) for i in starters]
# list of endings of voice detected segments (in frames)
enderFrames = [int(i) for i in enders]
# if the last voice detected segment isn't closed, use the last frame as end of this segment
if len(starterFrames) > len(enderFrames):
enderFrames.append(totalFrames)
# list of beginnings of voice detected segments (in milliseconds)
starterMs = [int((i / frameRate) * 1000) for i in starterFrames]
def run(self):
#global my_mutex
global numRecordings
global audioNum
global processNum
megaArray = []
currDay = 1
fileChar = "A"
wavFile = ""
#while processNum < numRecordings:
while goingFlag == True or processNum < audioNum:
while processNum >= audioNum:
y = 0
#print processNum
try:
if processNum > 0:
subprocess.check_output(['rm', wavFile])
#jsonFile = "/home/pi/results" + `processNum` + ".json"
wavFile = dir_path + "/talking" + ` processNum ` + ".wav"
print "Processing " + wavFile
#foFile = "fo" + `processNum` + ".fo"
#pmFile = "pm" + `processNum` + ".pm"
timeEnd = q.get()
tempDay = currDay
currDay = qSoundJson.get()
if tempDay != currDay:
megaArray = []
soundName = dir_path + "/data/day" + str(
currDay) + "/" + fileChar + "/sound" + str(
fileNum) + ".json"
if fileChar == "A":
fileChar = "B"
else:
fileChar = "A"
print timeEnd
#print subprocess.check_output(['python', '/home/pi/detectVoiceInWav.py', wavFile, jsonFile, str(timeEnd)])
v = VoiceActivityDetector(wavFile)
ampData = v._calculate_amplitude(v.data)
if np.percentile(ampData, 75) < 200:
print "AUTO"
speech_labels = []
speech_dict = {}
speech_dict['percent_time_w_speech'] = 0.0
speech_dict['time_started'] = round(timeEnd, 3)
speech_dict['length_of_recording'] = round(
lengthRecording, 3)
speech_dict['mean_amp'] = round(
float(float(sum(ampData)) / len(ampData)), 3)
speech_dict['med_amp'] = round(np.median(ampData), 3)
speech_dict['25_amp'] = round(np.percentile(ampData, 25),
3)
speech_dict['75_amp'] = round(np.percentile(ampData, 75),
3)
speech_dict['max_amp'] = round(np.amax(ampData), 3)
speech_dict['processed'] = 0
speech_labels.append(speech_dict)
print "AUTO DONE"
else:
print "MANUAL"
raw_detection = v.detect_speech()
#print raw_detection
speech_labels = v.convert_windows_to_readible_labels(
raw_detection, str(timeEnd))
print "MANUAL DONE"
megaArray.append(speech_labels)
save_to_file(megaArray, soundName)
#print subprocess.check_output(['./reaper/REAPER/build/reaper', '-i', wavFile, '-f', foFile, '-p', pmFile, '-a'])
processNum = processNum + 1
except subprocess.CalledProcessError as e:
logging.exception("message")
subprocess.check_output(['rm', wavFile])
sense.clear()
with open("processWav.txt", "w") as text_file:
text_file.write("Total wavs processed: {}".format(str(processNum)))
with open("createWav.txt", "w") as t2:
t2.write("Total wavs supposed to be processed: {}".format(
str(audioNum)))
currTime = time.time()
currTime = currTime - timeStart
with open("timeTaken.txt", "w") as t3:
t3.write("Time taken to process wavs: {}".format(str(currTime)))
i = 0
total_speech = 0
try:
while (True):
speech_labels = {}
if (i < 4):
i = i + 1
else:
i = 0
wav_name = 'file' + str(i) + '.wav'
create_wav(wav_name)
v = VoiceActivityDetector(wav_name)
raw_detection = v.detect_speech()
speech_labels, speech_in_wav = v.convert_windows_to_readible_labels(
raw_detection)
if (speech_in_wav > 0.5):
print('we have speech in ' + wav_name)
else:
print('we dont have speech in ' + wav_name)
output_txt = 'testingreal' + str(i) + '.txt'
total_speech = total_speech + speech_in_wav
save_to_file(speech_labels, output_txt)
except KeyboardInterrupt:
print('Total time since program started:' + str(total_speech))
print("Done")
from vad import VoiceActivityDetector filename = '/home/leferrae/Desktop/These/Kunwok/tour_nabagardi/20190723_085214.wav' v = VoiceActivityDetector(filename) v.plot_detected_speech_regions()
from vad import VoiceActivityDetector # import json # Input file audiofile = "abc.wav" plotaudiosignal = VoiceActivityDetector(audiofile) # Plotting the Audio signals, so that we can identify when it is having silence in audio plotaudiosignal.plot_detected_speech_regions() # # def save_to_file(data, file): # with open(file, 'w') as fp: # json.dump(data, fp) # # # # v = VoiceActivityDetector(filename) # raw_detection = v.detect_speech() # print(raw_detection) # speech_labels = v.convert_windows_to_readible_labels(raw_detection) # # save_to_file(speech_labels, "bala.json")
from vad import VoiceActivityDetector import sys v = VoiceActivityDetector(sys.argv[1]) raw_detection = v.detect_speech() speech_labels = v.convert_windows_to_readible_labels(raw_detection) print(speech_labels)