def _convert_mp3_to_wav(map3_file: str) -> typing.Union[str, None]:
assert map3_file.endswith(".mp3")
out_file = nlp.replace_file_name(map3_file, ".mp3", ".wav")
if os.path.exists(out_file):
return out_file
cmd = f"ffmpeg -i {map3_file} {out_file}"
if execute_cmd(cmd) == 0:
return out_file
return None
def _convert_flac_to_wav(flac_file: str) -> typing.Union[str, None]:
assert flac_file.endswith(".flac")
out_file = nlp.replace_file_name(flac_file, ".flac", ".wav")
if os.path.exists(out_file):
return out_file
cmd = f"sox {flac_file} {out_file}"
if nlp.execute_cmd(cmd) == 0:
return out_file
return None
def convert_to_16bits(wav_or_flac_file: str) -> typing.Union[str, None]:
file_ext = nlp.get_file_extension(wav_or_flac_file)
new_file = nlp.replace_file_name(wav_or_flac_file, f".{file_ext}",
".16bits.wav")
if os.path.exists(new_file):
return new_file
if nlp.execute_cmd(f"sox {wav_or_flac_file} -b 16 {new_file}") == 0:
return new_file
return None
def _convert_sph_to_wav(sph_file: str) -> typing.Union[str, None]:
assert sph_file.endswith(".sph")
out_file = nlp.replace_file_name(sph_file, ".sph", ".wav")
if os.path.exists(out_file):
return out_file
cmd = f"sox {sph_file} {out_file}"
if execute_cmd(cmd) == 0:
return out_file
cmd = f"sph2pipe -f rif {sph_file} {out_file}"
if execute_cmd(cmd) == 0:
return out_file
return None
def convert_to_standard_wav(
wav_or_flac_file: str) -> typing.Union[str, None]:
'''
:return: sample-width=2 Bytes, sample rating=16K.
'''
if wav_or_flac_file.endswith(".norm.wav"):
return wav_or_flac_file
file_ext = nlp.get_file_extension(wav_or_flac_file)
new_file = nlp.replace_file_name(wav_or_flac_file, f".{file_ext}",
".norm.wav")
if os.path.exists(new_file):
return new_file
if nlp.execute_cmd(f"sox {wav_or_flac_file} "
f"-b 16 -r 16000 {new_file}") == 0:
return new_file
return None
def preemphasize_wav(standard_wav_file: str):
assert standard_wav_file.endswith(".norm.wav")
new_file = nlp.replace_file_name(standard_wav_file, ".norm.wav",
".norm.amp.wav")
if os.path.exists(new_file):
return new_file
sample_rate, signal = wavfile.read(standard_wav_file)
assert sample_rate == 16000
pre_emphasis = 0.97
emphasized_signal = numpy.append(
signal[0], signal[1:] - pre_emphasis * signal[:-1])
amplified_signal = emphasized_signal * (32768 /
emphasized_signal.max())
wavfile.write(new_file, sample_rate,
amplified_signal.astype(numpy.int16))
return new_file
def segment_audio(flac_or_wav_file: str, time_segments: list,
dest_folder: str) -> typing.Iterator:
'''
time_segments: [(12,97, 18.89), (18.43, 27.77) ...] in seconds.
return: an iterator retuning a new segment file. If one time segment is
invalid, then the its corresponding segment file name is None.
'''
file_ext = nlp.get_file_extension(flac_or_wav_file)
assert file_ext in ["flac", "wav"]
assert os.path.exists(dest_folder)
base_name = os.path.basename(flac_or_wav_file)
audio = AudioSegment.from_file(flac_or_wav_file, file_ext)
duration = len(audio)
for file_id, (time_from, time_to) in enumerate(time_segments):
t_from = time_from * 1000
t_to = time_to * 1000
if not (0 <= t_from < t_to < duration):
print(f"WARN: {flac_or_wav_file} is not complete. "
f"Actual length: {duration / 1000} seconds, "
f"while time segment is {time_from}-{time_to}")
yield None
continue
seg_name = os.path.join(
dest_folder,
nlp.replace_file_name(base_name, f".{file_ext}",
f".{file_id:04}.{file_ext}"))
try:
audio[t_from:t_to].export(seg_name, format=file_ext)
yield seg_name
except Exception as error:
print_flush(error)
yield None