def generate_wavs(data_dir):
print("hello")
pbar = ProgressBar()
for mp3_file in pbar(glob(path.join(data_dir, '*.mp3'))):
sound = pydub.AudioSegment.from_mp3(mp3_file)
filename = mp3_file[-10:-4]
new_file = path.splitext(data_dir)[0] + "/wavs/" + filename + ".wav"
sound.export(new_file, format="wav")
pbar = ProgressBar()
data_dir = data_dir + '/wavs/'
# change audio file to 16k sample rate
for wav_file in pbar(glob(path.join(data_dir, '*.wav'))):
new_file = path.splitext(wav_file)[0] + "k16.wav"
transformer = Transformer()
transformer.convert(samplerate=sample_rate)
transformer.build(wav_file, new_file)
pbar = ProgressBar()
# remove old files
for item in pbar(glob(path.join(data_dir, '*.wav'))):
if item.endswith("k16.wav"):
continue
else:
os.remove(item)
pbar = ProgressBar()
# rename files to remove k16
for item in pbar(glob(path.join(data_dir, '*.wav'))):
os.rename(item, item.replace('k16', ''))
print("end")
def convert(self):
"""Converts the mp3's associated with this instance to wav's
Return:
wav_directory (os.path): The directory into which the associated wav's were downloaded
"""
wav_directory = self._pre_convert()
for mp3_filename in self.mp3_directory.glob('**/*.mp3'):
wav_filename = path.join(wav_directory, os.path.splitext(os.path.basename(mp3_filename))[0] + ".wav")
if not path.exists(wav_filename):
_logger.debug("Converting mp3 file %s to wav file %s" % (mp3_filename, wav_filename))
transformer = Transformer()
transformer.convert(samplerate=SAMPLE_RATE, n_channels=N_CHANNELS, bitdepth=BITDEPTH)
transformer.build(str(mp3_filename), str(wav_filename))
else:
_logger.debug("Already converted mp3 file %s to wav file %s" % (mp3_filename, wav_filename))
return wav_directory
def read(self, audio_metadata):
"""Read an audio file.
:param audio_metadata: metadata info of an audio
:return: raw audio data as float32 array and duration in seconds.
"""
fd = temp_path = None
# Convert it to a wav file.
if not audio_metadata.path.endswith('.wav'):
original_sample_rate = file_info.sample_rate(audio_metadata.path)
assert self._sample_rate <= original_sample_rate
transformer = Transformer()
transformer.convert(samplerate=self._sample_rate,
n_channels=self._channels,
bitdepth=self._bits_per_sample)
fd, temp_path = tempfile.mkstemp(suffix='.wav')
transformer.build(audio_metadata.path, temp_path)
if temp_path:
path = temp_path
else:
path = audio_metadata.path
# Read the audio file.
with SoundFile(path) as soundfile:
# make sure the audio properties are as expected.
assert soundfile.samplerate == self._sample_rate
assert soundfile.channels == self._channels
duration_sec = len(soundfile) / self._sample_rate
pcm = soundfile.read(dtype='float32')
# Add 0.5 second silence to the end of files containing keyword as in occasionally the user stopped
# recording right after uttering the keyword. If the detector needs some time after seeing the keyword to
# make a decision (e.g. endpointing) this is going to artificially increase the miss rates.
if audio_metadata.is_keyword:
pcm = np.append(pcm, np.zeros(self._sample_rate // 2))
if temp_path:
os.close(fd)
os.remove(temp_path)
return pcm, duration_sec
def convert(self):
"""Converts the mp3's associated with this instance to wav's
Return:
wav_directory (os.path): The directory into which the associated wav's were downloaded
"""
wav_directory = self._pre_convert()
for mp3_filename in self.mp3_directory.glob('**/*.mp3'):
wav_filename = path.join(
wav_directory,
os.path.splitext(os.path.basename(mp3_filename))[0] + ".wav")
if not path.exists(wav_filename):
_logger.debug("Converting mp3 file %s to wav file %s" %
(mp3_filename, wav_filename))
transformer = Transformer()
transformer.convert(samplerate=SAMPLE_RATE,
n_channels=N_CHANNELS,
bitdepth=BITDEPTH)
transformer.build(str(mp3_filename), str(wav_filename))
else:
_logger.debug("Already converted mp3 file %s to wav file %s" %
(mp3_filename, wav_filename))
return wav_directory
def _processSamples(sample_list):
for sample in sample_list:
sample_new_name = _renameSample(sample)
_out = join(out_path, sample_new_name)
processed_samples.append(_out)
_in = sample
# Sox processing using Transform instance
tfm = Transformer()
tfm.convert(samplerate=44100, n_channels=2, bitdepth=16)
if NORMALIZE:
tfm.norm(db_level=-3)
if SILENCE:
tfm.silence(location=-1,
silence_threshold=0.05,
min_silence_duration=0.1)
if PADDING:
tfm.pad(0, PADDING)
tfm.build(_in, _out)
def _maybe_convert_wav(mp3_filename, wav_filename):
if not path.exists(wav_filename):
transformer = Transformer()
transformer.convert(samplerate=SAMPLE_RATE)
transformer.build(mp3_filename, wav_filename)
def convert_sr_channel(audio) : transformer = Transformer() transformer.convert(samplerate=16000, n_channels=1) transformer.build(original_audio_file, wav_file)
