search

NVIDIA / waveglow

A Flow-based Generative Network for Speech Synthesis
BSD 3-Clause "New" or "Revised" License
2.29k stars 530 forks source link

Convert log Mel bank energy to audio by your model #263

Open zuujhyt opened 2 years ago

zuujhyt commented 2 years ago

Hi, I use a toolbox from python_speech_features import logfbank to get log mel energy from wav file, can I convert the feature back to speech b waveform bby your model? If yes, could you please guide? Thank you.

from python_speech_features import logfbank
from scipy.io import wavfile
sample_rate, wav_data = wavfile.read('test.wav')
audio_feats = logfbank(wav_data, samplerate=sample_rate).astype(np.float32)
##
# How to convert audio_feats back to wavform?
##