def prosody_static(self, audio, plots):
"""Extract the static prosody features from an audio file
:param audio: .wav audio file.
:param plots: timeshift to extract the features
:returns: array with the 103 prosody features
>>> prosody=Prosody()
>>> file_audio="../audios/001_ddk1_PCGITA.wav"
>>> features=prosody.prosody_static(file_audio, plots=True)
"""
fs, data_audio=read(audio)
data_audio=data_audio-np.mean(data_audio)
data_audio=data_audio/float(np.max(np.abs(data_audio)))
size_frameS=self.size_frame*float(fs)
size_stepS=self.step*float(fs)
thr_len_pause=self.thr_len*float(fs)
overlap=size_stepS/size_frameS
nF=int((len(data_audio)/size_frameS/overlap))-1
if self.pitch_method == 'praat':
name_audio=audio.split('/')
temp_uuid='prosody'+name_audio[-1][0:-4]
if not os.path.exists(self.PATH+'/../tempfiles/'):
os.makedirs(self.PATH+'/../tempfiles/')
temp_filename_f0=self.PATH+'/../tempfiles/tempF0'+temp_uuid+'.txt'
temp_filename_vuv=self.PATH+'/../tempfiles/tempVUV'+temp_uuid+'.txt'
praat_functions.praat_vuv(audio, temp_filename_f0, temp_filename_vuv, time_stepF0=self.step, minf0=self.minf0, maxf0=self.maxf0)
F0,_=praat_functions.decodeF0(temp_filename_f0,len(data_audio)/float(fs),self.step)
os.remove(temp_filename_f0)
os.remove(temp_filename_vuv)
elif self.pitch_method == 'rapt':
data_audiof=np.asarray(data_audio*(2**15), dtype=np.float32)
F0=pysptk.sptk.rapt(data_audiof, fs, int(size_stepS), min=self.minf0, max=self.maxf0, voice_bias=self.voice_bias, otype='f0')
segmentsV=V_UV(F0, data_audio, fs, type_seg="Voiced", size_stepS=size_stepS)
segmentsUP=V_UV(F0, data_audio, fs, type_seg="Unvoiced", size_stepS=size_stepS)
segmentsP=[]
segmentsU=[]
for k in range(len(segmentsUP)):
eu=logEnergy(segmentsUP[k])
if (len(segmentsUP[k])>thr_len_pause):
segmentsP.append(segmentsUP[k])
else:
segmentsU.append(segmentsUP[k])
F0_features=F0feat(F0)
energy_featuresV=energy_feat(segmentsV, fs, size_frameS, size_stepS)
energy_featuresU=energy_feat(segmentsU, fs, size_frameS, size_stepS)
duration_features=duration_feat(segmentsV, segmentsU, segmentsP, data_audio, fs)
if plots:
self.plot_pros(data_audio,fs,F0,segmentsV, segmentsU, F0_features)
features=np.hstack((F0_features, energy_featuresV, energy_featuresU, duration_features))
return features
def prosody_static(audio, flag_plots):
fs, data_audio = read(audio)
data_audio = data_audio[:-1, 0]
print(len(data_audio))
print(data_audio)
data_audio = data_audio - np.mean(data_audio)
data_audio = data_audio / float(np.max(np.abs(data_audio)))
size_frameS = 0.02 * float(fs)
size_stepS = 0.01 * float(fs)
thr_len_pause = 0.14 * float(fs)
thr_en_pause = 0.2
overlap = size_stepS / size_frameS
nF = int((len(data_audio) / size_frameS / overlap)) - 1
data_audiof = np.asarray(data_audio * (2**15), dtype=np.float32)
print(data_audiof)
print(size_stepS)
F0 = pysptk.sptk.rapt(data_audiof,
fs,
int(size_stepS),
min=60,
max=350,
voice_bias=-0.2,
otype='f0')
logE = []
for l in range(nF):
data_frame = data_audio[int(l * size_stepS):int(l * size_stepS +
size_frameS)]
logE.append(logEnergy(data_frame))
logE = np.asarray(logE)
print("see")
print(np.unique(F0))
segmentsV = V_UV(F0,
data_audio,
fs,
type_seg="Voiced",
size_stepS=size_stepS)
segmentsU = V_UV(F0,
data_audio,
fs,
type_seg="Unvoiced",
size_stepS=size_stepS)
Nvoiced = len(segmentsV)
Nunvoiced = len(segmentsU)
Vrate = fs * float(Nvoiced) / len(data_audio)
avgdurv = 1000 * np.mean([len(segmentsV[k])
for k in range(Nvoiced)]) / float(fs)
stddurv = 1000 * np.std([len(segmentsV[k])
for k in range(Nvoiced)]) / float(fs)
silence = []
for k in range(Nunvoiced):
eu = logEnergy(segmentsU[k])
if (eu < thr_en_pause or len(segmentsU[k]) > thr_len_pause):
silence.append(segmentsU[k])
print("here")
print(eu)
Silrate = fs * float(len(silence)) / len(data_audio)
avgdurs = 1000 * np.mean([len(silence[k])
for k in range(len(silence))]) / float(fs)
stddurs = 1000 * np.std([len(silence[k])
for k in range(len(silence))]) / float(fs)
if flag_plots:
plt.figure(1)
plt.subplot(311)
t = np.arange(0, float(len(data_audio)) / fs, 1.0 / fs)
if len(t) != len(data_audio):
t = np.arange(1.0 / fs, float(len(data_audio)) / fs, 1.0 / fs)
print(len(t), len(data_audio))
plt.plot(t, data_audio, 'k')
plt.ylabel('Amplitude')
plt.xlabel('Time (s)')
plt.xlim([0, t[-1]])
plt.grid(True)
plt.subplot(312)
fsp = len(F0) / t[-1]
print(fsp)
t2 = np.arange(0.0, t[-1], 1.0 / fsp)
if len(t2) > len(F0):
t2 = t2[:len(F0)]
elif len(F0) > len(t2):
F0 = F0[:len(t2)]
plt.plot(t2, F0, color='k', linewidth=2.0)
plt.xlabel('Time (s)')
plt.ylabel('F0 (Hz)')
plt.ylim([0, np.max(F0) + 10])
plt.xlim([0, t[-1]])
plt.grid(True)
plt.subplot(313)
fse = len(logE) / t[-1]
t3 = np.arange(0.0, t[-1], 1.0 / fse)
if len(t3) > len(logE):
t3 = t3[:len(logE)]
elif len(logE) > len(t3):
logE = logE[:len(t3)]
plt.plot(t3, logE, color='k', linewidth=2.0)
plt.xlabel('Time (s)')
plt.ylabel('Energy (dB)')
#plt.ylim([0,np.max(logE)])
plt.xlim([0, t[-1]])
plt.grid(True)
plt.show()
F0std = np.std(F0[F0 != 0])
F0varsemi = Hz2semitones(F0std**2)
return F0, logE, np.mean(F0[F0 != 0]), np.std(
F0[F0 != 0]), np.max(F0), np.mean(logE), np.std(logE), np.max(
logE
), Vrate, avgdurv, stddurv, Silrate, avgdurs, stddurs, F0varsemi
def prosody_static(audio, flag_plots, pitch_method='praat'):
fs, data_audio = read(audio)
data_audio = data_audio - np.mean(data_audio)
data_audio = data_audio / float(np.max(np.abs(data_audio)))
size_frameS = 0.02 * float(fs)
size_stepS = 0.01 * float(fs)
thr_len_pause = 0.14 * float(fs)
thr_en_pause = 10 * np.log10(0.02)
overlap = size_stepS / size_frameS
nF = int((len(data_audio) / size_frameS / overlap)) - 1
if pitch_method == 'praat':
temp_uuid = audio.split('/')[-1][0:-4]
temp_filename_f0 = path_app + '/../tempfiles/tempF0' + temp_uuid + '.txt'
temp_filename_vuv = '../tempfiles/tempVUV' + temp_uuid + '.txt'
praat_functions.praat_vuv(audio,
temp_filename_f0,
temp_filename_vuv,
time_stepF0=0.01,
minf0=60,
maxf0=350)
F0, _ = praat_functions.decodeF0(temp_filename_f0,
len(data_audio) / float(fs), 0.01)
os.remove(temp_filename_f0)
elif pitch_method == 'rapt':
data_audiof = np.asarray(data_audio * (2**15), dtype=np.float32)
F0 = pysptk.sptk.rapt(data_audiof,
fs,
int(size_stepS),
min=60,
max=350,
voice_bias=-0.2,
otype='f0')
segmentsV = V_UV(F0,
data_audio,
fs,
type_seg="Voiced",
size_stepS=size_stepS)
segmentsUP = V_UV(F0,
data_audio,
fs,
type_seg="Unvoiced",
size_stepS=size_stepS)
segmentsP = []
segmentsU = []
for k in range(len(segmentsUP)):
eu = logEnergy(segmentsUP[k])
if (len(segmentsUP[k]) > thr_len_pause):
segmentsP.append(segmentsUP[k])
else:
segmentsU.append(segmentsUP[k])
F0_features = F0feat(F0)
energy_featuresV = energy_feat(segmentsV, fs, size_frameS, size_stepS)
energy_featuresU = energy_feat(segmentsU, fs, size_frameS, size_stepS)
duration_features = duration_feat(segmentsV, segmentsU, segmentsP,
data_audio, fs)
if flag_plots:
plot_pros(data_audio, fs, F0, segmentsV, segmentsU)
features = np.hstack(
(F0_features, energy_featuresV, energy_featuresU, duration_features))
return features