Tacotron-2 icon indicating copy to clipboard operation
Tacotron-2 copied to clipboard

Published 20 hours ago • verified publisher icon Rayhane-mamah

275 seconds per step

Open ErfolgreichCharismatisch opened this issue 6 years ago • 4 comments

  • 203 custom audio files
  • 2 to 15 seconds long
  • customized metadata.csv Wavefile|content|contentcopy
  • I changed outputs_per_step to 6 in hparams.py
  • I have a Geforce GTX 960 with CUDA

this is my current output

ausgabe

275 seconds per step is slow. Can it be accelerated?

ErfolgreichCharismatisch avatar Sep 10 '18 20:09 ErfolgreichCharismatisch

I use 1080 Ti for training 7 sec to 12 sec samples only takes me 1.5 sec/step. Maybe you didn't use your GPU?

lkfo415579 avatar Sep 12 '18 01:09 lkfo415579

You have made some changes in hparams. Do you think that is the reason?

Have you ever tried using as default?

Thien223 avatar Sep 12 '18 02:09 Thien223

Maybe you didn't use your GPU?

You were right. Now I have a OOM problem: https://github.com/Rayhane-mamah/Tacotron-2/issues/196

ErfolgreichCharismatisch avatar Sep 12 '18 10:09 ErfolgreichCharismatisch

@ErfolgreichCharismatisch how were your results? Can you share audio samples? I have a similar sized dataset and wondering whether it will work for trying to clone a speaker

vivekkalyanarangan30 avatar Oct 30 '20 17:10 vivekkalyanarangan30