Text-to-Speech
Fairseq
English
audio
Changhan's picture
Update config.yaml
203a320
raw
history blame
621 Bytes
features:
energy_max: 3.2244551181793213
energy_min: -4.9544901847839355
eps: 1.0e-05
f_max: 8000
f_min: 0
hop_len_t: 0.011609977324263039
hop_length: 256
n_fft: 1024
n_mels: 80
n_stft: 513
pitch_max: 5.733940816898645
pitch_min: -4.660287183665281
sample_rate: 22050
type: spectrogram+melscale+log
win_len_t: 0.046439909297052155
win_length: 1024
window_fn: hann
global_cmvn:
stats_npz_path: fbank_mfa_gcmvn_stats.npz
transforms:
'*':
- global_cmvn
vocab_filename: vocab.txt
vocoder:
type: hifigan
config: hifigan.json
checkpoint: hifigan.bin
inference:
tokenization: g2p