unilight/seq2seq-vc

Conversion of own dataset using Pretrained models

Opened this issue · 1 comments

MohanThippeswamy commented

I tried conversion (lsc decoding) on own dataset (10 files) using pretrained models.
Results are something bla bla...bla..

I followed below steps

  1. s3prl-vc-extract-upstream -successful
  2. compute_staticts -skipped
  3. normalize -failed
  4. vc_decode. (used raw feature's from step-1)
    Please let me know some pointer to look at..
MohanThippeswamy commented

@unilight,
I was able to decode for my own dataset.
After successful execution of normalization step it worked flawless and got to see good conversion. But minor change in the content of source audio is observed.
Thanks