Releases: kan-bayashi/ParallelWaveGAN
Releases · kan-bayashi/ParallelWaveGAN
Version 0.4.5
- Simplify decoding part
- Add
load_model
function - Add
inference
method in each genearator - Add
download_pretraiend_model
function to download from google drive directly
Version 0.4.3
- Fix PQMF #204
Version 0.4.2
- Fixed wrong loss naming
- Fixed shuffle in distributed training
- Fixed pad mode
- Updated to be able to change PQMF paramters
Version 0.4.1
- Fix numba version issue
- Support multi-key hdf5 scp format
- Support npy scp format
Version 0.4.0
What's new
- Support Multi-band MelGAN
- Support Full-band MelGAN
Version 0.3.5
Version 0.3.4
What's new
Scripts
- Support
--pretrain
option in training - Support
--skip-wav-copy
option in normalization - Support
scp
style input for all/bin
scripts for ESPnet compatibility - Better parallelization (much faster in the case of the large dataset)
Recipes
- Fix
format: npy
support - Add VCTK recipe
- Add
melgan.v3
config - Add
parallel_wavegan.v3
config
Version 0.3.0
What's new
- Support more recipes
- Support new residual discriminator
- Support MelGAN generator
- Support MelGAN discriminator
- And more refactoring...
Version 0.2.5
Stable version release.