geneing / WaveRNN

Pytorch implementation of Deepmind's WaveRNN model
13 stars 1 forks source link
pytorch-implementation speech-to-text text-to-speech vocoder wavenet-vocoder wavernn

WaveRNN

This code is forked from https://github.com/fatchord/WaveRNN and optimized for Mozilla-TTS.

Released Models

Models Dataset Commit Audio Sample TTS model Details
mold model LJspeech 8a1c152 soundcloud Tacotron2-iter-260K Model with Mixture of Logistic Distribution
10bit model LJSpeech faea90b soundcloud Tacotron2-iter-260K 10bit Softmax output
universal vocoder LibriTTS 12c8744 soundcloud - (details)

Check this TTS notebook to see TTS+WaveRNN in action. To train your own model, you can use ExtractTTSSpectrogram to generate spectrograms by TTS and train WaveRNN. It might be also interesting to check this TTS issue to catchup with the current state.