Python pyworld.get_cheaptrick_fft_size方法代码示例

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def create_synthesizer(
            self,
            buffer_size: int,
            number_of_pointers: int,
    ):
        assert self._synthesizer is None

        self._synthesizer = structures.WorldSynthesizer()
        apidefinitions._InitializeSynthesizer(
            self.out_sampling_rate,  # sampling rate
            self.acoustic_param.frame_period,  # frame period
            pyworld.get_cheaptrick_fft_size(self.out_sampling_rate),  # fft size
            buffer_size,  # buffer size
            number_of_pointers,  # number of pointers
            self._synthesizer,
        ) 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def __call__(self, data: Wave, test):
        acoustic_feature = self._acoustic_feature_process(data, test=True).astype_only_float(self._dtype)
        high_spectrogram = acoustic_feature.spectrogram

        fftlen = pyworld.get_cheaptrick_fft_size(data.sampling_rate)
        low_spectrogram = pysptk.mc2sp(
            acoustic_feature.mfcc,
            alpha=self._alpha,
            fftlen=fftlen,
        )

        feature = LowHighSpectrogramFeature(
            low=low_spectrogram,
            high=high_spectrogram,
        )
        feature.validate()
        return feature 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def __init__(
            self,
            acoustic_feature_param: AcousticFeatureParam,
            out_sampling_rate: int,
            buffer_size: int,
            number_of_pointers: int,
    ):
        from world4py.native import structures, apidefinitions
        super().__init__(
            acoustic_feature_param=acoustic_feature_param,
            out_sampling_rate=out_sampling_rate,
        )

        self.buffer_size = buffer_size

        self._synthesizer = structures.WorldSynthesizer()
        apidefinitions._InitializeSynthesizer(
            self.out_sampling_rate,  # sampling rate
            self.acoustic_feature_param.frame_period,  # frame period
            pyworld.get_cheaptrick_fft_size(out_sampling_rate),  # fft size
            buffer_size,  # buffer size
            number_of_pointers,  # number of pointers
            self._synthesizer,
        )
        self._before_buffer = []  # for holding memory 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def world_decode_spectral_envelop(coded_sp, fs):

    fftlen = pyworld.get_cheaptrick_fft_size(fs)
    #coded_sp = coded_sp.astype(np.float32)
    #coded_sp = np.ascontiguousarray(coded_sp)
    decoded_sp = pyworld.decode_spectral_envelope(coded_sp, fs, fftlen)

    return decoded_sp 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def get_sizes(sampling_rate: int, order: int):
        fft_size = pyworld.get_cheaptrick_fft_size(fs=sampling_rate)
        return dict(
            f0=1,
            spectrogram=fft_size // 2 + 1,
            aperiodicity=fft_size // 2 + 1,
            mfcc=order + 1,
            voiced=1,
        ) 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def generate_file(path):
    out = Path(arguments.output_directory, path.stem + '.npy')
    if out.exists() and not arguments.enable_overwrite:
        return

    # load wave and padding
    wave_file_load_process = WaveFileLoadProcess(
        sample_rate=arguments.sample_rate,
        top_db=arguments.top_db,
        pad_second=arguments.pad_second,
    )
    wave = wave_file_load_process(path, test=True)

    # make acoustic feature
    acoustic_feature_process = AcousticFeatureProcess(
        frame_period=arguments.frame_period,
        order=arguments.order,
        alpha=arguments.alpha,
        f0_estimating_method=arguments.f0_estimating_method,
    )
    feature = acoustic_feature_process(wave, test=True).astype_only_float(numpy.float32)
    high_spectrogram = feature.spectrogram

    fftlen = pyworld.get_cheaptrick_fft_size(arguments.sample_rate)
    low_spectrogram = pysptk.mc2sp(
        feature.mfcc,
        alpha=arguments.alpha,
        fftlen=fftlen,
    )

    # save
    numpy.save(out.absolute(), {
        'low': low_spectrogram,
        'high': high_spectrogram,
    }) 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def world_decode_spectral_envelop(coded_sp, fs):
    fftlen = pyworld.get_cheaptrick_fft_size(fs)
    # coded_sp = coded_sp.astype(np.float32)
    # coded_sp = np.ascontiguousarray(coded_sp)
    decoded_sp = pyworld.decode_spectral_envelope(coded_sp, fs, fftlen)

    return decoded_sp 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def decode_spectrogram(self, feature: AcousticFeature):
        fftlen = pyworld.get_cheaptrick_fft_size(self.out_sampling_rate)
        feature.sp = pysptk.mc2sp(
            feature.mc.astype(numpy.float32),
            alpha=pysptk.util.mcepalpha(self.out_sampling_rate),
            fftlen=fftlen,
        )
        return feature 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def get_sizes(sampling_rate: int, order: int):
        fft_size = pyworld.get_cheaptrick_fft_size(fs=sampling_rate)
        return dict(
            f0=1,
            sp=fft_size // 2 + 1,
            ap=fft_size // 2 + 1,
            mc=order + 1,
            mc_wop=order,
            voiced=1,
        ) 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def mc2sp(mc: numpy.ndarray, sampling_rate: int, alpha: float):
        return pysptk.mc2sp(
            mc.astype(numpy.float64),
            alpha=alpha,
            fftlen=pyworld.get_cheaptrick_fft_size(sampling_rate),
        ) 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def decode_ap(ap: numpy.ndarray, sampling_rate: int):
        return pyworld.decode_aperiodicity(
            ap.astype(numpy.float64),
            sampling_rate,
            pyworld.get_cheaptrick_fft_size(sampling_rate),
        ) 

# 需要导入模块: import pyworld [as 别名]
# 或者: from pyworld import get_cheaptrick_fft_size [as 别名]
def convert_to_feature(self, input: AcousticFeature, out_sampling_rate: Optional[int] = None):
        if out_sampling_rate is None:
            out_sampling_rate = self.config.dataset.param.voice_param.sample_rate

        input_feature = input
        input = self._feature_normalize(input, test=True)
        input = self._encode_feature(input, test=True)

        pad = 128 - input.shape[1] % 128
        input = numpy.pad(input, [(0, 0), (0, pad)], mode='minimum')

        converter = partial(chainer.dataset.convert.concat_examples, device=self.gpu, padding=0)
        inputs = converter([input])

        with chainer.using_config('train', False):
            out = self.model(inputs).data[0]

        if self.gpu is not None:
            out = chainer.cuda.to_cpu(out)
        out = out[:, :-pad]

        out = self._decode_feature(out, test=True)
        out = AcousticFeature(
            f0=out.f0,
            spectrogram=out.spectrogram,
            aperiodicity=out.aperiodicity,
            mfcc=out.mfcc,
            voiced=input_feature.voiced,
        )
        out = self._feature_denormalize(out, test=True)
        out = AcousticFeature(
            f0=out.f0,
            spectrogram=out.spectrogram,
            aperiodicity=input_feature.aperiodicity,
            mfcc=out.mfcc,
            voiced=out.voiced,
        )

        fftlen = pyworld.get_cheaptrick_fft_size(out_sampling_rate)
        spectrogram = pysptk.mc2sp(
            out.mfcc,
            alpha=self._param.acoustic_feature_param.alpha,
            fftlen=fftlen,
        )

        out = AcousticFeature(
            f0=out.f0,
            spectrogram=spectrogram,
            aperiodicity=out.aperiodicity,
            mfcc=out.mfcc,
            voiced=out.voiced,
        ).astype(numpy.float64)
        return out