def decode(
self,
acoustic_feature: AcousticFeature,
):
length = len(acoustic_feature.f0)
f0_buffer = utils.cast_1d_list_to_1d_pointer(acoustic_feature.f0.flatten().tolist())
sp_buffer = utils.cast_2d_list_to_2d_pointer(acoustic_feature.spectrogram.tolist())
ap_buffer = utils.cast_2d_list_to_2d_pointer(acoustic_feature.aperiodicity.tolist())
apidefinitions._AddParameters(f0_buffer, length, sp_buffer, ap_buffer, self._synthesizer)
ys = []
while apidefinitions._Synthesis2(self._synthesizer) != 0:
y = numpy.array([self._synthesizer.buffer[i] for i in range(self.buffer_size)])
ys.append(y)
if len(ys) > 0:
out_wave = Wave(
wave=numpy.concatenate(ys),
sampling_rate=self.out_sampling_rate,
)
else:
out_wave = Wave(
wave=numpy.empty(0),
sampling_rate=self.out_sampling_rate,
)
self._before_buffer.append((f0_buffer, sp_buffer, ap_buffer)) # for holding memory
if len(self._before_buffer) > 16:
self._before_buffer.pop(0)
return out_wave
def decode(
self,
acoustic_feature: AcousticFeature,
):
from world4py.native import apidefinitions, utils
length = len(acoustic_feature.f0)
f0_buffer = utils.cast_1d_list_to_1d_pointer(acoustic_feature.f0.flatten().tolist())
sp_buffer = utils.cast_2d_list_to_2d_pointer(acoustic_feature.spectrogram.tolist())
ap_buffer = utils.cast_2d_list_to_2d_pointer(acoustic_feature.aperiodicity.tolist())
apidefinitions._AddParameters(f0_buffer, length, sp_buffer, ap_buffer, self._synthesizer)
ys = []
while apidefinitions._Synthesis2(self._synthesizer) != 0:
y = numpy.array([self._synthesizer.buffer[i] for i in range(self.buffer_size)])
ys.append(y)
if len(ys) > 0:
out_wave = Wave(
wave=numpy.concatenate(ys),
sampling_rate=self.out_sampling_rate,
)
else:
out_wave = Wave(
wave=numpy.empty(0),
sampling_rate=self.out_sampling_rate,
)
self._before_buffer.append((f0_buffer, sp_buffer, ap_buffer)) # for holding memory
if len(self._before_buffer) > 16:
self._before_buffer.pop(0)
return out_wave
def code_aperiodicity(aperiodicity, fs, f0_length, fft_size):
'''Compress aperiodicity's dimension
Args:
aperiodicity (POINTER(POINTER(c_double))): Aperiodicity
fs (int): Sampling frequency [Hz]
f0_length (int): Rank of aperiodicity 1d
fft_size (int): FFT size
Returns:
POINTER(POINTER(c_double)): compressed aperiodicity
Notice:
You can get compressed apeiodicity's rank via `get_codec_aperiodicity` method
'''
tmp_2d_array = [[0 for i in range(get_codec_aperiodicity_num(fs))]
for o in range(f0_length)]
coded_aperiodicity = utils.cast_2d_list_to_2d_pointer(
tmp_2d_array, ctypes.c_double)
apidefinitions._CodeAperiodicity(aperiodicity, f0_length, fs, fft_size,
coded_aperiodicity)
return code_aperiodicity
def decode_spectral_envelope(coded_spectral_envelope, fs, f0_length,
number_of_dimensions, fft_size):
'''Restore compressed spectral envelope's dimension
Args:
coded_spectral_envelope (POINTER(POINTER(c_double))): Coded spectral envelope
fs (int): Sampling frequency [Hz]
f0_length (int): Rank of spectrogram 1d
number_of_dimensions (int): Number of compressed spectral envelope's dimension
fft_size (int): FFT size
Returns:
POINTER(POINTER(c_double)): Spectral envelope
'''
tmp_2d_array = [[0 for i in range(number_of_dimensions)]
for o in range(f0_length)]
spectrogram = utils.cast_2d_list_to_2d_pointer(tmp_2d_array,
ctypes.c_double)
apidefinitions._DecodeSpectralEnvelope(coded_spectral_envelope, f0_length,
fs, fft_size, number_of_dimensions,
spectrogram)
return spectrogram
def d4c(x,
fs,
temporal_positions,
f0,
threshold=0.85,
f0_floor=DEFAULT_F0_FLOOR,
fft_size=None,
**dummy):
'''Aperiodicity estimation by D4C
Args:
x (POINTER(c_double)): Input waveform
fs (int): a
temporal_positions (POINTER(c_double)): Temporal positions
f0 (POINTER(c_double)): Extracted F0
threshold (double, optional): To determine frame is unvoice or not
(high value is tending to regard as unvoiced)
f0_floor (double, optional): To determine fft_size
fft_size (double, optional): When fft_size is not set, determined by fs and default f0_floor
Returns:
POINTER(POINTER(c_double)): Aperiodicity
Tuple(int, int): Aperiodicity's shape "(f0_length, fft_size // 2 + 1)"
'''
option = structures.D4COption()
apidefinitions._InitializeD4COption(option)
option.threshold = threshold
if fft_size is None:
tmp_fft_size = get_fft_size(fs, f0_floor)
else:
tmp_fft_size = fft_size
f0_length = get_f0_length(f0)
tmp_2d_array = [[0 for i in range(tmp_fft_size // 2 + 1)]
for o in range(f0_length)]
aperiodicity = utils.cast_2d_list_to_2d_pointer(tmp_2d_array,
ctypes.c_double)
apidefinitions._D4C(x, len(list(x)), fs, temporal_positions, f0, f0_length,
tmp_fft_size, option, aperiodicity)
return aperiodicity, tmp_fft_size, (f0_length, tmp_fft_size // 2 + 1)
def cheap_trick(x,
fs,
temporal_positions,
f0,
q1=-0.15,
f0_floor=DEFAULT_F0_FLOOR,
fft_size=None,
**dummy):
'''Spectral envelope estimation by CheapTrick
Args:
x (POINTER(c_double)): Input waveform
fs (int): Sampling frequency [Hz] (To determine fft_size)
temporal_positions (POINTER(c_double)): Temporal positions
f0 (POINTER(c_double)): Extracted F0
q1 (double, optional): Spectral recovery parameter (should not change)
f0_floor (double, optional): To determine fft_size
fft_size (int, optional): When fft_size is not set, determined by fs and default f0_floor
Returns:
POINTER(POINTER(c_double)): Spectral envelope
int: FFT size used by CheapTrick
Tuple(int, int): Spectrogram's shape "(f0_length, fft_size // 2 + 1)"
'''
option = structures.CheapTrickOption()
apidefinitions._InitializeCheapTrickOption(fs, option)
option.q1 = q1
if fft_size is None:
option.f0_floor = f0_floor
option.fft_size = apidefinitions._GetFFTSizeForCheapTrick(fs, option)
else:
option.fft_size = fft_size
f0_length = get_f0_length(f0)
tmp_2d_array = [[0 for i in range(option.fft_size // 2 + 1)]
for o in range(f0_length)]
spectrogram = utils.cast_2d_list_to_2d_pointer(tmp_2d_array,
ctypes.c_double)
apidefinitions._CheapTrick(x, len(list(x)), fs, temporal_positions, f0,
f0_length, option, spectrogram)
return spectrogram, option.fft_size, (f0_length, option.fft_size // 2 + 1)
def decode_aperiodicity(coded_aperiodicity, fs, f0_length, fft_size):
'''Restore compressed aperiodicity's dimension
Args:
coded_aperiodicity (POINTER(POINTER(c_double))): Coded aperiodicity
fs (int): Sampling frequency [Hz]
f0_length (int): Rank of aperiodicity 1d
fft_size (int): FFT size
Returns:
POINTER(POINTER(c_double)): Aperiodicity
'''
tmp_2d_array = [[0 for i in range(get_codec_aperiodicity_num(fs))]
for o in range(f0_length)]
aperiodicity = utils.cast_2d_list_to_2d_pointer(tmp_2d_array,
ctypes.c_double)
apidefinitions._DecodeAperiodicity(coded_aperiodicity, f0_length, fs,
fft_size, aperiodicity)
return aperiodicity