Uberduck

well maybe it errored at 28.wav because it did them in a random order? not sure

best to put them through the ffmpeg grinder just in case

```
!apt install parallel >&-
!find "/content/wavs" -type f -name "*.wav" -print | parallel --eta --bar ffmpeg -i "{'\x7b\x7d'}" -ac 1 -ar 22050 -c:a pcm_s16le -loglevel error -f wav "{'\x7b.\x7d'}.wa2"
!find "/content/wavs" -type f -name "*.wav" -delete
!find "/content/wavs" -type f -name "*.wa2" -print | rename s/\.wa2/\.wav/
```

so breaths before a sentance or in middle?

cause ive been doing the sentence in audacity ending it with a breath

at the beginning or end would mess with synthesis negatively

How do I associate a tuned hifigan with an uploaded voice?

Does Anyone Know How To Remove Music From Videos Over 7 Minutes?

So I have been training some voices but I've been noticing that "validation" only appears in some voices, in some it's missing entirely and would only have the "train" part (e.x., I should get both Loss/train and Loss/val, but the latter is missing sometimes). Is there a problem with something in my end? (Pipeline)

I heard that you stop it once the graph only starts plateu (meaning it doesn't go down anymore)

I'm pretty new to this training stuff and I don't even see a Loss/val graph in my end sometimes but that's what I heard

once the graph starts plateau i may start training in hifigan is that right?

Unless Tensorboard doesn't work then look at loss

for torchmoji override you're not supposed to put in actual emoji, but rather text that would be analyzed to have the proper emotion
but also torchmoji doesn't really work even on very large datasets