Browse Source

README.md: updating readme to explicitly mention that mel representation of WaveNet and Tacotron2 must be the same

master
rafaelvalle 6 years ago
parent
commit
7eb045206c
1 changed files with 4 additions and 0 deletions
  1. +4
    -0
      README.md

+ 4
- 0
README.md View File

@ -38,9 +38,13 @@ wavenet](https://github.com/r9y9/wavenet_vocoder/)
1. `python -m multiproc train.py --output_directory=outdir --log_directory=logdir --hparams=distributed_run=True,fp16_run=True`
## Inference
When performing Mel-Spectrogram to Audio synthesis with a WaveNet model, make sure Tacotron 2 and WaveNet were trained on the same mel-spectrogram representation. Follow these steps to use a a simple inference pipeline using griffin-lim:
1. `jupyter notebook --ip=127.0.0.1 --port=31337`
2. load inference.ipynb
## Related repos
[nv-wavenet](https://github.com/NVIDIA/nv-wavenet/): Faster than real-time
wavenet inference

Loading…
Cancel
Save