Skip to content

ishine/Mutiband-HIFIGAN

Repository files navigation

Mutiband-HIFIGAN

An Easy Implementation of HIFIGAN with muit-band.

This project is based on:

Something new

  • Support different datasets
  • Support muti-band
  • Support different sample_rate, hop_size ...

Current Training...

Firstly, the datasets structure like below.

datasets:
    wavs:
        speaker1:
            ???.wav
        speaker2:
            ???.wav
        ..... 

Secondly, create training.txt

python preprocess.py

Thirdly, training

python train.py

Some problems

while using pqmf merge four subbands(value between -1 and 1), value may greater than 1.0 and less than -1.0. I think it may cause by PQMF filters (can't reconstruct perfectly)

About

Mutiband version of HIFIGAN

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages