unilight/seq2seq-vc

How to get pre-trained models for cascade and stg methods

MohanThippeswamy opened this issue · 2 comments

@unilight ,
Please help me get pretrained models for cascade and STG method.

Thanks
Mohan

@unilight,
Wanted to check availability of pretrained models for cascade & STG method

Currently going with models downloaded in stage -1.
Stage -1: Data and pre-trained model download.
Make sure L2-ARCTIC is downloaded to somewhere locally and properly set in run.sh.
The CMU ARCTIC dataset is downloaded (by default to downloads/).
A pre-trained vocoder is downloaded (to downloads/pwg_TXHC). We use a Parallel WaveGAN vocoder.
The non-parallel frame-based model is downloaded (to downloads/s3prl-vc-ppg_sxliu).
The pre-trained seq2seq VC model is downloaded (to downloads/ljspeech_transformer_tts_aept).