def __init__(self, config: dict):
super().__init__(config)
self.spect = Spectrum(config)
self.cmvn = CMVN(config)
# global cmvn dim == feature dim
if config.type == 'MelSpectrum' and self.cmvn.global_cmvn:
assert config.filterbank_channel_count * config.channel == len(config.global_mean), \
'Error, feature dim {} is not equals to cmvn dim {}'. \
format(config.filterbank_channel_count * config.channel,
len(config.global_mean))
def __init__(self, config: dict):
super().__init__(config)
self.spect = Spectrum(config)
self.cmvn = CMVN(config)
# global cmvn dim == feature dim
if config.type == "Fbank" and self.cmvn.global_cmvn:
assert config.filterbank_channel_count * config.channel == len(
config.global_mean
), "Error, feature dim {} is not equals to cmvn dim {}".format(
config.filterbank_channel_count * config.channel,
len(config.global_mean),
)
print("Fbank params: ", self.config)
def test_spectrum(self):
wav_path_16k = str(
Path(os.environ["MAIN_ROOT"]).joinpath("examples/sm1_cln.wav"))
wav_path_8k = str(
Path(os.environ["MAIN_ROOT"]).joinpath("examples/english.wav"))
with self.session():
for wav_file in [wav_path_8k, wav_path_16k]:
read_wav = ReadWav.params().instantiate()
input_data, sample_rate = read_wav(wav_file)
spectrum = Spectrum.params({
"window_length": 0.025,
"dither": 0.0
}).instantiate()
spectrum_test = spectrum(input_data, sample_rate)
output_true = np.array([
[9.819611, 2.84503, 3.660894, 2.7779, 1.212233],
[9.328745, 2.553949, 3.276319, 3.000918, 2.499342],
])
if tf.executing_eagerly():
self.assertEqual(tf.rank(spectrum_test).numpy(), 2)
else:
self.assertEqual(tf.rank(spectrum_test).eval(), 2)
if wav_file == wav_path_16k:
if tf.executing_eagerly():
self.assertAllClose(
spectrum_test.numpy()[0:2, 0:5],
output_true,
rtol=1e-05,
atol=1e-05,
)
else:
self.assertAllClose(
spectrum_test.eval()[0:2, 0:5],
output_true,
rtol=1e-05,
atol=1e-05,
)
def params(cls, config=None):
"""
Set params.
:param config: contains thirteen optional parameters:upper_frequency_limit(float, default=0),
lower_frequency_limit(float, default=60.0), filterbank_channel_count(float, default=40.0),
:return: An object of class HParams, which is a set of hyperparameters as name-value pairs.
"""
hparams = HParams(cls=cls)
# spectrum
hparams.append(Spectrum.params({"output_type": 1, "is_fbank": True}))
# fbank
upper_frequency_limit = 0
lower_frequency_limit = 60
filterbank_channel_count = 40
hparams.add_hparam("upper_frequency_limit", upper_frequency_limit)
hparams.add_hparam("lower_frequency_limit", lower_frequency_limit)
hparams.add_hparam("filterbank_channel_count",
filterbank_channel_count)
# delta
delta_delta = False # True
order = 2
window = 2
hparams.add_hparam("delta_delta", delta_delta)
hparams.add_hparam("order", order)
hparams.add_hparam("window", window)
if config is not None:
hparams.parse(config, True)
hparams.type = "Fbank"
hparams.add_hparam("channel", 1)
if hparams.delta_delta:
hparams.channel = hparams.order + 1
return hparams
def params(cls, config=None):
"""
Set params.
:param config: contains thirteen optional parameters:
--window_length : Window length in seconds. (float, default = 0.025)
--frame_length : Hop length in seconds. (float, default = 0.010)
--snip_edges : If True, the last frame (shorter than window_length) will be
cutoff. If False, 1 // 2 frame_length data will be padded
to data. (bool, default = True)
---raw_energy : If 1, compute frame energy before preemphasis and
windowing. If 2, compute frame energy after
preemphasis and windowing. (int, default = 1)
--preEph_coeff : Coefficient for use in frame-signal preemphasis.
(float, default = 0.0)
--window_type : Type of window ("hamm"|"hann"|"povey"|"rect"|"blac"|"tria").
(string, default = "hann")
--remove_dc_offset : Subtract mean from waveform on each frame.
(bool, default = false)
--is_fbank : If true, compute power spetrum without frame energy.
If false, using the frame energy instead of the
square of the constant component of the signal.
(bool, default = true)
--output_type : If 1, return power spectrum. If 2, return log-power
spectrum. If 3, return magnitude spectrum. (int, default = 3)
--upper_frequency_limit : High cutoff frequency for mel bins (if <= 0, offset
from Nyquist) (float, default = 0)
--lower_frequency_limit : Low cutoff frequency for mel bins (float, default = 20)
--filterbank_channel_count : Number of triangular mel-frequency bins.
(float, default = 23)
--dither : Dithering constant (0.0 means no dither).
(float, default = 0) [add robust to training]
:return: An object of class HParams, which is a set of hyperparameters as name-value pairs.
"""
hparams = HParams(cls=cls)
# spectrum
hparams.append(
Spectrum.params({
'output_type': 3,
'is_fbank': True,
'preEph_coeff': 0.0,
'window_type': 'hann',
'dither': 0.0,
'remove_dc_offset': False
}))
# mel_spectrum
upper_frequency_limit = 0
lower_frequency_limit = 60
filterbank_channel_count = 40
sample_rate = -1
hparams.add_hparam('upper_frequency_limit', upper_frequency_limit)
hparams.add_hparam('lower_frequency_limit', lower_frequency_limit)
hparams.add_hparam('filterbank_channel_count',
filterbank_channel_count)
hparams.add_hparam('sample_rate', sample_rate)
# delta
delta_delta = False # True
order = 2
window = 2
hparams.add_hparam('delta_delta', delta_delta)
hparams.add_hparam('order', order)
hparams.add_hparam('window', window)
if config is not None:
hparams.parse(config, True)
hparams.type = 'MelSpectrum'
hparams.add_hparam('channel', 1)
if hparams.delta_delta:
hparams.channel = hparams.order + 1
return hparams
def params(cls, config=None):
"""Set params.
Args:
config: contains the following thirteen optional parameters:
'window_length': Window length in seconds. (float, default = 0.025)
'frame_length': Hop length in seconds. (float, default = 0.010)
'snip_edges': If 1, the last frame (shorter than window_length) will be
cutoff. If 2, 1 // 2 frame_length data will be padded
to data. (int, default = 1)
'preEph_coeff': Coefficient for use in frame-signal preemphasis.
(float, default = 0.97)
'window_type': Type of window ("hamm"|"hann"|"povey"|"rect"|"blac"|"tria").
(string, default = "povey")
'remove_dc_offset': Subtract mean from waveform on each frame.
(bool, default = true)
'is_fbank': If true, compute power spetrum without frame energy.
If false, using the frame energy instead of the
square of the constant component of the signal.
(bool, default = true)
'is_log10': If true, using log10 to fbank. If false, using loge.
(bool, default = false)
'output_type': If 1, return power spectrum. If 2, return log-power
spectrum. (int, default = 1)
'upper_frequency_limit': High cutoff frequency for mel bins (if <= 0, offset
from Nyquist) (float, default = 0)
'lower_frequency_limit': Low cutoff frequency for mel bins (float, default = 20)
'filterbank_channel_count': Number of triangular mel-frequency bins.
(float, default = 23)
'dither': Dithering constant (0.0 means no dither).
(float, default = 1) [add robust to training]
Note:
Return an object of class HParams, which is a set of hyperparameters as
name-value pairs.
"""
hparams = HParams(cls=cls)
# spectrum
hparams.append(Spectrum.params({"output_type": 1, "is_fbank": True}))
# fbank
upper_frequency_limit = 0
lower_frequency_limit = 60
filterbank_channel_count = 40
is_log10 = False
hparams.add_hparam("upper_frequency_limit", upper_frequency_limit)
hparams.add_hparam("lower_frequency_limit", lower_frequency_limit)
hparams.add_hparam("filterbank_channel_count",
filterbank_channel_count)
hparams.add_hparam('is_log10', is_log10)
# delta
delta_delta = False # True
order = 2
window = 2
hparams.add_hparam("delta_delta", delta_delta)
hparams.add_hparam("order", order)
hparams.add_hparam("window", window)
if config is not None:
hparams.parse(config, True)
hparams.type = "Fbank"
hparams.add_hparam("channel", 1)
if hparams.delta_delta:
hparams.channel = hparams.order + 1
return hparams