athena.tools.vocoder¶
Implementation of GriffinLim vocoder
Module Contents¶
Classes¶
python implementation of griffinlim algorithm |
- class athena.tools.vocoder.GriffinLim(data_descriptions)¶
python implementation of griffinlim algorithm
- _get_nfft(window_length)¶
n_fft is an exponential power of 2 closest to and larger than win_length
- __call__(feats, hparams, name=None)¶
- _logmel_to_linear(feats)¶
Convert FBANK to linear spectrogram. :param feats: FBANK feats, shape: [length, channels]
- Returns
Linear spectrogram
- Return type
linear_feats
- _griffin_lim(linear_feats, gl_iters)¶
Convert linear spectrogram into waveform
- Parameters
linear_feats – linear spectrogram
gl_iters – num of gl iterations
- Returns
Reconstructed waveform (N,).
- Return type
waveform