def _record(self):
if self.isrecording:
CHUNK = 1024
FORMAT = pyaudio.paInt16
CHANNELS = 1
RATE = 16000
WAVE_OUTPUT_FILENAME = "rec.wav"
p = pyaudio.PyAudio()
stream = p.open(format=FORMAT,
channels=CHANNELS,
rate=RATE,
input=True,
frames_per_buffer=CHUNK)
self.play = True
print("* recording")
frames = []
i = 1
while self.isrecording:
self.button.config(image=self.photo[i],
width="325",
height="200")
data = stream.read(CHUNK)
frames.append(data)
i = i + 1
if (i == 10):
i = 0
self.button.config(image=self.photo[0], width="325", height="200")
print("* done recording")
stream.stop_stream()
stream.close()
p.terminate()
wf = wave.open(WAVE_OUTPUT_FILENAME, 'wb')
wf.setnchannels(CHANNELS)
wf.setsampwidth(p.get_sample_size(FORMAT))
wf.setframerate(RATE)
wf.writeframes(b''.join(frames))
wf.close()
wav_file = 'rec.wav'
feat_file = 'rec.feat'
x, s = sf.read(wav_file)
fe = sp.FrontEnd(samp_rate=16000, mean_norm_feat=True)
feat = fe.process_utterance(x)
htk.write_htk_user_feat(feat, feat_file)
feat_rscp_line = os.path.join(feat_file)
line = "%s=%s[0,%d]\n" % (feat_file, feat_rscp_line,
feat.shape[1] - 1)
try:
ret = sd_beta.decode(line, args.trn, args.beam_width,
args.lmweight, z, fst, 1)
self.ques.set(ret)
nlp_main()
except KeyboardInterrupt:
print('End')
# plot waveform
plt.plot(x)
plt.title('waveform')
plt.savefig('fig/waveform.png', bbox_inches='tight')
plt.close()
# plot mel filterbank
for i in range(0, fe.num_mel):
plt.plot(fe.mel_filterbank[i, :])
plt.title('mel filterbank')
plt.savefig('fig/mel_filterbank.png', bbox_inches='tight')
plt.close()
# plot log mel spectrum (fbank)
plt.imshow(
feat, origin='lower', aspect=4
) # flip the image so that vertical frequency axis goes from low to high
plt.title('log mel filterbank features (fbank)')
plt.savefig('fig/fbank.png', bbox_inches='tight')
plt.close()
htk.write_htk_user_feat(feat, feat_file)
print("Wrote {0} frames to {1}".format(feat.shape[1], feat_file))
#
#if you want to verify, that the file was written correctly:
feat2 = htk.read_htk_user_feat(name=feat_file)
print("Read {0} frames rom {1}".format(feat2.shape[1], feat_file))
print("Per-element absolute error is {0}".format(
np.linalg.norm(feat - feat2) / (feat2.shape[0] * feat2.shape[1])))
