def call(self, audio_data, sample_rate=None):
"""
Caculate mfcc features of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: A float tensor of size (num_channels, num_frames, num_frequencies) containing
mfcc features of every frame in speech.
"""
p = self.config
with tf.name_scope('mfcc'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
spectrum_feats = self.spect(audio_data, sample_rate)
spectrum_feats = tf.expand_dims(spectrum_feats, 0)
fbank_feats = self.fbank(audio_data, sample_rate)
mfcc = py_x_ops.mfcc(fbank_feats,
spectrum_feats,
sample_rate,
use_energy=p.use_energy,
cepstral_lifter=p.cepstral_lifter,
coefficient_count=p.coefficient_count)
return mfcc
def call(self, audio_data, sample_rate=None):
"""
Caculate cepstrum of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return:A float tensor of size (num_frames, ceps_subband_num) containing normalized cepstrum
(tag_ceps_mean_norm = True) or cepstrum (tag_ceps_mean_norm = False) of every frame in speech.
"""
p = self.config
with tf.name_scope('cepstrum'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=float)
assert_op = tf.assert_equal(
tf.constant(p.sample_rate), tf.cast(sample_rate, dtype=float))
with tf.control_dependencies([assert_op]):
cepstrum = py_x_ops.cepstrum(
audio_data,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length,
ceps_subband_num=p.ceps_subband_num,
tag_ceps_mean_norm=p.tag_ceps_mean_norm)
return cepstrum
def call(self, audio_data, sample_rate=None):
"""
Caculate power spectrum and phase spectrum of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: Two returns:
power spectrum —— A float tensor of size (num_frames, num_frequencies) containing
power spectrum and of every frame in speech.
phase spectrum —— A float tensor of size (num_frames, num_frequencies) containing
phase spectrum and of every frame in speech.
"""
p = self.config
with tf.name_scope('analyfiltbank'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
sample_rate = tf.cast(sample_rate, dtype=float)
power_spectrum, phase_spectrum = py_x_ops.analyfiltbank(
audio_data,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length)
return power_spectrum, phase_spectrum
def call(self, audio_data, sample_rate=None):
"""
Caculate power spectrum or log power spectrum of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: A float tensor of size (num_frames, num_frequencies) containing power spectrum (output_type=1)
or log power spectrum (output_type=2) of every frame in speech.
"""
p = self.config
with tf.name_scope('spectrum'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
sample_rate = tf.cast(sample_rate, dtype=float)
spectrum = py_x_ops.spectrum(
audio_data,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length,
output_type=p.output_type,
snip_edges=p.snip_edges,
raw_energy=p.raw_energy,
preEph_coeff=p.preeph_coeff,
window_type=p.window_type,
remove_dc_offset=p.remove_dc_offset,
is_fbank=p.is_fbank)
return spectrum
def call(self, audio_data, sample_rate=None):
"""
Caculate power spectrum or log power spectrum of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: A float tensor of size N containing add-noise audio.
"""
p = self.config
with tf.name_scope('add_rir_noise_aecres'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(
tf.constant(p.sample_rate), tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
sample_rate = tf.cast(sample_rate, dtype=float)
add_rir_noise_aecres_out = py_x_ops.add_rir_noise_aecres(
audio_data,
sample_rate,
if_add_rir=p.if_add_rir,
rir_filelist=p.rir_filelist,
if_add_noise=p.if_add_noise,
snr_min=p.snr_min,
snr_max=p.snr_max,
noise_filelist=p.noise_filelist,
if_add_aecres=p.if_add_aecres,
aecres_filelist=p.aecres_filelist)
return tf.squeeze(add_rir_noise_aecres_out)
def call(self, audio_data, sample_rate=None):
"""
Caculate fbank && pitch(concat) features of wav.
:param audio_data: the audio signal from which to compute spectrum.
Should be an (1, N) tensor.
:param sample_rate: the samplerate of the signal we working with.
:return: A tensor with shape (num_frames, dim_features), containing
fbank && pitch feature of every frame in speech.
"""
p = self.config
with tf.name_scope('fbank_pitch'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
fbank_feats = tf.squeeze(self.fbank(audio_data, sample_rate))
pitch_feats = tf.squeeze(self.pitch(audio_data, sample_rate))
fbank_pitch_feats = tf.concat([fbank_feats, pitch_feats], 1)
return fbank_pitch_feats
def call(self, power_spectrum, phase_spectrum, sample_rate=None):
"""
Implement frequency domain to time domain conversion.
:param power_spectrum: a float tensor of size (num_frames, num_frequencies).
:param phase_spectrum: a float tensor of size (num_frames, num_frequencies).
:param sample_rate: a scalar tensor.
:return: audio data
"""
p = self.config
with tf.name_scope('synthfiltbank'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
audio_data = py_x_ops.synthfiltbank(
power_spectrum,
phase_spectrum,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length)
return audio_data
def call(self, audio_data, sample_rate=None):
"""
Caculate fbank features of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: A float tensor of size (num_channels, num_frames, num_frequencies) containing
fbank features of every frame in speech.
"""
p = self.config
with tf.name_scope('fbank'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
if p.upper_frequency_limit <= 0:
p.upper_frequency_limit = p.sample_rate / 2.0 + p.upper_frequency_limit
elif (p.upper_frequency_limit <= p.lower_frequency_limit) or (
p.upper_frequency_limit > p.sample_rate / 2.0):
p.upper_frequency_limit = p.sample_rate / 2.0
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
spectrum = self.spect(audio_data, sample_rate)
spectrum = tf.expand_dims(spectrum, 0)
fbank = py_x_ops.fbank(
spectrum,
sample_rate,
upper_frequency_limit=p.upper_frequency_limit,
lower_frequency_limit=p.lower_frequency_limit,
filterbank_channel_count=p.filterbank_channel_count)
return fbank
def call(self, audio_data, sample_rate=None):
"""
Calculate the zero-crossing rate of speech.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: A tensor with shape (1, num_frames), containing zero-crossing rate of every frame in speech.
"""
p = self.config
with tf.name_scope('zcr'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
sample_rate = tf.cast(sample_rate, dtype=float)
zcr = py_x_ops.zcr(audio_data,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length)
return zcr
def call(self, audio_data, sample_rate=None):
"""
Caculate power of every frame in speech.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return:A float tensor of size (1, num_frames) containing power of every frame in speech.
"""
p = self.config
with tf.name_scope('framepow'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=float)
assert_op = tf.assert_equal(
tf.constant(p.sample_rate), tf.cast(sample_rate, dtype=float))
with tf.control_dependencies([assert_op]):
framepow = py_x_ops.frame_pow(
audio_data,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length)
return framepow
def call(self, audio_data, sample_rate=None):
"""
Caculate plp features of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return:A float tensor of size (num_frames, (plp_order + 1)) containing plp features of every frame in speech.
"""
p = self.config
with tf.name_scope('plp'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
sample_rate = tf.cast(sample_rate, dtype=float)
plp = py_x_ops.plp(audio_data,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length,
plp_order=p.plp_order)
return plp
def accuracy(logits, labels):
''' accuracy candies
params:
logits: [B, ..., D]
labels: [B, ...]
return:
accuracy tensor
'''
with tf.name_scope('accuracy'):
assert_rank = tf.assert_equal(tf.rank(logits), tf.rank(labels) + 1)
assert_shape = tf.assert_equal(tf.shape(logits)[:-1], tf.shape(labels))
with tf.control_dependencies([assert_rank, assert_shape]):
predictions = tf.argmax(logits, axis=-1, output_type=tf.int64)
labels = tf.cast(labels, tf.int64)
return tf.reduce_mean(
tf.cast(tf.equal(predictions, labels), dtype=tf.float32))
def call(self, audio_data, sample_rate=None):
"""
Caculate pitch features of audio data.
:param audio_data: the audio signal from which to compute spectrum. Should be an (1, N) tensor.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: A float tensor of size (1, num_frames) containing pitch features of every frame in speech.
"""
p = self.config
with tf.name_scope('pitch'):
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=float)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=float))
with tf.control_dependencies([assert_op]):
pitch = py_x_ops.pitch(audio_data,
sample_rate,
window_length=p.window_length,
frame_length=p.frame_length,
thres_autoc=p.thres_autoc)
pitch = tf.squeeze(pitch)
pitch = tf.transpose(pitch[None, :])
return pitch
def call(self, wavfile):
"""
Get audio data and sample rate from a wavfile.
:param wavfile: filepath of wav
:return: 2 values. The first is a Tensor of audio data. The second return value is the sample rate of the input wav
file, which is a tensor with float dtype.
"""
p = self.config
contents = tf.io.read_file(wavfile)
audio_data, sample_rate = tf.audio.decode_wav(
contents, desired_channels=p.audio_channels)
assert_op = tf.assert_equal(tf.constant(p.sample_rate),
tf.cast(sample_rate, dtype=float))
with tf.control_dependencies([assert_op]):
return tf.squeeze(audio_data, axis=-1), tf.cast(sample_rate,
dtype=float)
def call(self, filename, audio_data, sample_rate=None):
"""
Write wav using audio_data[tensor].
:param filename: filepath of wav.
:param audio_data: a tensor containing data of a wav.
:param sample_rate: [option]the samplerate of the signal we working with, default is 16kHz.
:return: write wav opration.
"""
p = self.config
filename = tf.constant(filename)
if sample_rate == None:
sample_rate = tf.constant(p.sample_rate, dtype=tf.int32)
assert_op = tf.assert_equal(
tf.constant(p.sample_rate), tf.cast(sample_rate, dtype=tf.int32))
with tf.control_dependencies([assert_op]):
audio_data = tf.cast(audio_data, dtype=tf.float32)
contents = tf.audio.encode_wav(
tf.expand_dims(audio_data, 1), tf.cast(sample_rate, dtype=tf.int32))
w = tf.io.write_file(filename, contents)
return w
