def _adjust_time_resolution(self, batch, local_condition, max_time_steps):
'''Adjust time resolution between audio and local condition
'''
if local_condition:
new_batch = []
for b in batch:
x, c, g, l = b
self._assert_ready_for_upsample(x, c)
if max_time_steps is not None:
max_steps = _ensure_divisible(
max_time_steps, audio.get_hop_size(self._hparams),
True)
if len(x) > max_time_steps:
max_time_frames = max_steps // audio.get_hop_size(
self._hparams)
start = np.random.randint(0, len(c) - max_time_frames)
time_start = start * audio.get_hop_size(self._hparams)
x = x[time_start:time_start + max_time_frames *
audio.get_hop_size(self._hparams)]
c = c[start:start + max_time_frames, :]
self._assert_ready_for_upsample(x, c)
new_batch.append((x, c, g, l))
return new_batch
else:
new_batch = []
for b in batch:
x, c, g, l = b
x = audio.trim_silence(x, hparams)
if max_time_steps is not None and len(x) > max_time_steps:
start = np.random.randint(0, len(c) - max_time_steps)
x = x[start:start + max_time_steps]
new_batch.append((x, c, g, l))
return new_batch
def _process_utterance(out_dir, index, wav_path, text):
'''Preprocesses a single utterance audio/text pair.
This writes the mel and linear scale spectrograms to disk and returns a tuple to write
to the train.txt file.
Args:
out_dir: The directory to write the spectrograms into
index: The numeric index to use in the spectrogram filenames.
wav_path: Path to the audio file containing the speech input
text: The text spoken in the input audio file
Returns:
A (spectrogram_filename, mel_filename, n_frames, text) tuple to write to train.txt
'''
# text to pinyin
text = text.replace("#1", "").replace("#2",
"").replace("#3",
"").replace("#4", "")
pinyin = " ".join(get_pinyin(text))
# Load the audio to a numpy array:
wav = audio.load_wav(wav_path)
wav = wav / np.max(np.abs(wav)) * 0.9 # norm
# denoise
if hparams.mmse_denoise_by_bothEndOfAudio and len(
wav) > hparams.sample_rate * (hparams.length_as_noise * 2 + 0.1):
noise_wav = np.concatenate([
wav[:int(hparams.sample_rate * hparams.length_as_noise)],
wav[-int(hparams.sample_rate * hparams.length_as_noise):]
])
profile = logmmse.profile_noise(noise_wav, hparams.sample_rate)
wav = logmmse.denoise(wav, profile, eta=0)
# trim silence
wav = audio.trim_silence(
wav, hparams.trim_top_db) # top_db=30 for aishell, 60 for BZNSYP
# audio.save_wav(wav, wav_path.replace(".wav", "_trimed.wav"))
# convert wav to 16bit int
wav *= 32768
wav = wav.astype(np.int16)
# extract LPC feature
extractor = lpcnet.FeatureExtractor()
feat = extractor.compute_feature(wav)
n_frames = feat.shape[0]
# write the lpc feature to disk
feature_filename = 'biaobei-lpc-feat-%05d.npy' % index
np.save(os.path.join(out_dir, feature_filename), feat, allow_pickle=False)
# Return a tuple describing this training example:
return (feature_filename, n_frames, pinyin)
def _process_utterance(out_dir, index, wav_path, pinyin):
'''Preprocesses a single utterance audio/text pair.
This writes the mel and linear scale spectrograms to disk and returns a tuple to write
to the train.txt file.
Args:
out_dir: The directory to write the spectrograms into
index: The numeric index to use in the spectrogram filenames.
wav_path: Path to the audio file containing the speech input
pinyin: The pinyin of Chinese spoken in the input audio file
Returns:
A (spectrogram_filename, mel_filename, n_frames, text) tuple to write to train.txt
'''
# Load the audio to a numpy array:
wav = audio.load_wav(wav_path)
# rescale wav for unified measure for all clips
wav = wav / np.abs(wav).max() * 0.999
# trim silence
wav = audio.trim_silence(wav)
# Compute the linear-scale spectrogram from the wav:
spectrogram = audio.spectrogram(wav).astype(np.float32)
n_frames = spectrogram.shape[1]
if n_frames > hp.max_frame_num:
return None
# Compute a mel-scale spectrogram from the wav:
mel_spectrogram = audio.melspectrogram(wav).astype(np.float32)
# Write the spectrograms to disk:
spectrogram_filename = 'thchs30-spec-%05d.npy' % index
mel_filename = 'thchs30-mel-%05d.npy' % index
np.save(os.path.join(out_dir, spectrogram_filename),
spectrogram.T,
allow_pickle=False)
np.save(os.path.join(out_dir, mel_filename),
mel_spectrogram.T,
allow_pickle=False)
# Return a tuple describing this training example:
return (spectrogram_filename, mel_filename, n_frames, pinyin)
def _process_utterance(out_dir, name, wav_path, text, hparams):
'''Preprocesses a single utterance audio/text pair.
This writes the mel and linear scale spectrograms to disk and returns a tuple to write
to the train.txt file.
Args:
out_dir: The directory to write the spectrograms into
index: The numeric index to use in the spectrogram filenames.
wav_path: Path to the audio file containing the speech input
text: The text spoken in the input audio file
Returns:
A (spectrogram_filename, mel_filename, n_frames, text) tuple to write to train.txt
'''
# Load the audio to a numpy array:
wav = audio.load_wav(wav_path, hparams)
# trim silences here
wav = audio.trim_silence(wav, hparams)
# Compute the linear-scale spectrogram from the wav:
spectrogram = audio.spectrogram(wav, hparams).astype(np.float32)
n_frames = spectrogram.shape[1]
# Compute a mel-scale spectrogram from the wav:
mel_spectrogram = audio.melspectrogram(wav, hparams).astype(np.float32)
# Write the spectrograms to disk:
spectrogram_filename = 'mailabs-spec-{}.npy'.format(name)
mel_filename = 'mailabs-mel-{}.npy'.format(name)
np.save(os.path.join(out_dir, spectrogram_filename),
spectrogram.T,
allow_pickle=False)
np.save(os.path.join(out_dir, mel_filename),
mel_spectrogram.T,
allow_pickle=False)
# Return a tuple describing this training example:
return (spectrogram_filename, mel_filename, n_frames, text)
def _process_utterance(out_dir, index, wav_path, pinyin):
# 读取语音
wav = audio.load_wav(wav_path)
wav = wav / np.abs(wav).max() * 0.999
# 消除静音
wav = audio.trim_silence(wav)
# 得到语音的线性频谱和梅尔频谱
spectrogram = audio.spectrogram(wav).astype(np.float32)
n_frames = spectrogram.shape[1]
mel_spectrogram = audio.melspectrogram(wav).astype(np.float32)
# 保存两种频谱
spectrogram_filename = 'biaobei-spec-%05d.npy' % index
mel_filename = 'biaobei-mel-%05d.npy' % index
np.save(os.path.join(out_dir, spectrogram_filename),
spectrogram.T,
allow_pickle=False)
np.save(os.path.join(out_dir, mel_filename),
mel_spectrogram.T,
allow_pickle=False)
return (spectrogram_filename, mel_filename, n_frames, pinyin)
def _process_utterance(mel_dir, linear_dir, wav_dir, index, wav_path, text,
hparams):
"""
Preprocesses a single utterance wav/text pair
this writes the mel scale spectogram to disk and return a tuple to write
to the train.txt file
Args:
- mel_dir: the directory to write the mel spectograms into
- linear_dir: the directory to write the linear spectrograms into
- wav_dir: the directory to write the preprocessed wav into
- index: the numeric index to use in the spectogram filename
- wav_path: path to the audio file containing the speech input
- text: text spoken in the input audio file
- hparams: hyper parameters
Returns:
- A tuple: (audio_filename, mel_filename, linear_filename, time_steps, mel_frames, linear_frames, text)
"""
try:
# Load the audio as numpy array
wav = audio.load_wav(wav_path)
except FileNotFoundError: #catch missing wav exception
print(
'file {} present in csv metadata is not present in wav folder. skipping!'
.format(wav_path))
return None
# M-AILABS extra silence specific
wav = audio.trim_silence(wav)
#Pre-emphasize
wav = audio.preemphasis(wav)
#rescale wav
#wav = wav / np.abs(wav).max() * hparams.rescaling_max
#Assert all audio is in [-1, 1]
if (wav > 1.).any() or (wav < -1.).any():
#raise RuntimeError('wav has invalid value: {}'.format(wav))
print('file {} has invalid value. skipping!'.format(wav_path))
return None
#[-1, 1]
out = wav
constant_values = 0.
out_dtype = np.float32
# Compute the mel scale spectrogram from the wav
mel_spectrogram = audio.melspectrogram(wav).astype(np.float32)
mel_frames = mel_spectrogram.shape[1]
if mel_frames > hparams.max_frame_num:
return None
#Compute the linear scale spectrogram from the wav
linear_spectrogram = audio.spectrogram(wav).astype(np.float32)
linear_frames = linear_spectrogram.shape[1]
#sanity check
assert linear_frames == mel_frames
#Ensure time resolution adjustement between audio and mel-spectrogram
l_pad, r_pad, hop_size = audio.librosa_pad_lr(wav)
#Reflect pad audio signal on the right (Just like it's done in Librosa to avoid frame inconsistency)
out = np.pad(out, (l_pad, r_pad),
mode='constant',
constant_values=constant_values)
assert len(out) >= mel_frames * hop_size
#time resolution adjustement
#ensure length of raw audio is multiple of hop size so that we can use
#transposed convolution to upsample
out = out[:mel_frames * hop_size]
assert len(out) % hop_size == 0
time_steps = len(out)
# Write the spectrogram and audio to disk
audio_filename = 'audio-{}.npy'.format(index)
mel_filename = 'mel-{}.npy'.format(index)
linear_filename = 'linear-{}.npy'.format(index)
np.save(os.path.join(wav_dir, audio_filename),
out.astype(out_dtype),
allow_pickle=False)
np.save(os.path.join(mel_dir, mel_filename),
mel_spectrogram.T,
allow_pickle=False)
np.save(os.path.join(linear_dir, linear_filename),
linear_spectrogram.T,
allow_pickle=False)
# Return a tuple describing this training example
return (audio_filename, mel_filename, linear_filename, time_steps,
mel_frames, text)
def _process_utterance(mel_dir, linear_dir, wav_dir, index, wav_path, text):
"""
Preprocesses a single utterance wav/text pair
this writes the mel scale spectogram to disk and return a tuple to write
to the train.txt file
Args:
- mel_dir: the directory to write the mel spectograms into
- linear_dir: the directory to write the linear spectrograms into
- wav_dir: the directory to write the preprocessed wav into
- index: the numeric index to use in the spectogram filename
- wav_path: path to the audio file containing the speech input
- text: text spoken in the input audio file
- hparams: hyper parameters
Returns:
- A tuple: (audio_filename, mel_filename, linear_filename, time_steps, mel_frames, linear_frames, text)
"""
try:
# Load the audio as numpy array
wav = audio.load_wav(wav_path)
except FileNotFoundError: #catch missing wav exception
print(
'file {} present in csv metadata is not present in wav folder. skipping!'
.format(wav_path))
return None
#rescale wav
if hparams.rescale:
wav = wav / np.abs(wav).max() * hparams.rescaling_max
#M-AILABS extra silence specific
if hparams.trim_silence:
wav = audio.trim_silence(wav)
#Mu-law quantize
if is_mulaw_quantize(hparams.input_type):
#[0, quantize_channels)
out = mulaw_quantize(wav, hparams.quantize_channels)
#Trim silences
start, end = audio.start_and_end_indices(out,
hparams.silence_threshold)
wav = wav[start:end]
out = out[start:end]
constant_values = mulaw_quantize(0, hparams.quantize_channels)
out_dtype = np.int16
elif is_mulaw(hparams.input_type):
#[-1, 1]
out = mulaw(wav, hparams.quantize_channels)
constant_values = mulaw(0., hparams.quantize_channels)
out_dtype = np.float32
else:
#[-1, 1]
out = wav
constant_values = 0.
out_dtype = np.float32
# Compute the mel scale spectrogram from the wav
mel_spectrogram = audio.melspectrogram(wav).astype(np.float32)
mel_frames = mel_spectrogram.shape[1]
if mel_frames > hparams.max_mel_frames and hparams.clip_mels_length:
return None
#Compute the linear scale spectrogram from the wav
linear_spectrogram = audio.linearspectrogram(wav).astype(np.float32)
linear_frames = linear_spectrogram.shape[1]
#sanity check
assert linear_frames == mel_frames
#Ensure time resolution adjustement between audio and mel-spectrogram
fft_size = hparams.n_fft if hparams.win_size is None else hparams.win_size
l, r = audio.pad_lr(wav, fft_size, audio.get_hop_size())
#Zero pad for quantized signal
out = np.pad(out, (l, r), mode='constant', constant_values=constant_values)
assert len(out) >= mel_frames * audio.get_hop_size()
#time resolution adjustement
#ensure length of raw audio is multiple of hop size so that we can use
#transposed convolution to upsample
out = out[:mel_frames * audio.get_hop_size()]
assert len(out) % audio.get_hop_size() == 0
time_steps = len(out)
# Write the spectrogram and audio to disk
audio_filename = 'speech-audio-{:05d}.npy'.format(index)
mel_filename = 'speech-mel-{:05d}.npy'.format(index)
linear_filename = 'speech-linear-{:05d}.npy'.format(index)
np.save(os.path.join(wav_dir, audio_filename),
out.astype(out_dtype),
allow_pickle=False)
np.save(os.path.join(mel_dir, mel_filename),
mel_spectrogram.T,
allow_pickle=False)
np.save(os.path.join(linear_dir, linear_filename),
linear_spectrogram.T,
allow_pickle=False)
# Return a tuple describing this training example
return (audio_filename, mel_filename, linear_filename, time_steps,
mel_frames, text)