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')
def process_feat(samp_rate=48000):
wav_files = glob("dev/*/*.wav")
strings = np.unique(
pd.Series(wav_files).replace(
to_replace=r"\.\.\/[a-z0-9]+\/dev\/[a-zA-Z0-9]+\/",
value="",
regex=True).values.flatten()).tolist()
for ii, wav_file in tqdm(enumerate(wav_files)):
x, s = librosa.core.load(wav_file)
fe = sp.FrontEnd(samp_rate=s,
mean_norm_feat=True,
frame_duration=0.032,
frame_shift=0.02274,
hi_freq=8000)
feat = fe.process_utterance(x)
if strings[ii].split("-")[6].replace(".wav", "") not in feat_actors:
feat_actors[strings[ii].split("-")[6].replace(".wav", "")] = []
feat_actors[strings[ii].split("-")[6].replace(".wav", "")].append(feat)
return feat_actors
if not os.path.isfile(wav_file):
raise RuntimeError(
'input wav file is missing. Have you downloaded the LibriSpeech corpus?'
)
if not os.path.exists(os.path.join(data_dir, 'feat')):
os.mkdir(os.path.join(data_dir, 'feat'))
samp_rate = 16000
x, s = sf.read(wav_file)
if (s != samp_rate):
raise RuntimeError("LibriSpeech files are 16000 Hz, found {0}".format(s))
fe = sp.FrontEnd(samp_rate=samp_rate, mean_norm_feat=True)
feat = fe.process_utterance(x)
if (plot_output):
if not os.path.exists('fig'):
os.mkdir('fig')
# 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):
compute_stats = True
else:
compute_stats = False
wav_list = os.path.join(data_dir, "lists/wav_{0}.list".format(args.set))
feat_list = os.path.join(data_dir, "lists/feat_{0}.rscp".format(args.set))
feat_dir = os.path.join(data_dir, "feat")
rscp_dir = "..." # note ... is CNTK notation for "relative to the location of the list of feature files
mean_file = os.path.join(data_dir, "am/feat_mean.ascii")
invstddev_file = os.path.join(data_dir, "am/feat_invstddev.ascii")
wav_dir = "../.."
if not os.path.exists(os.path.join(data_dir, 'am')):
os.mkdir(os.path.join(data_dir, 'am'))
samp_rate = 16000
fe = sp.FrontEnd(samp_rate=samp_rate,
mean_norm_feat=True,
compute_stats=compute_stats)
# read lines
with open(wav_list) as f:
wav_files = f.readlines()
wav_files = [x.strip() for x in wav_files]
if not os.path.exists(feat_dir):
os.makedirs(feat_dir)
if not os.path.exists(os.path.dirname(feat_list)):
os.makedirs(os.path.dirname(feat_list))
out_list = open(feat_list, "w")
count = 0
for line in wav_files:
