search

152334H / tortoise-tts-fast

Fast TorToiSe inference (5x or your money back!)
GNU Affero General Public License v3.0
771 stars 179 forks source link

Crackling sound when using dpm++2m #24

Open rikabi89 opened 1 year ago

rikabi89 commented 1 year ago

Compare : ddim : https://vocaroo.com/1ccLP3IZFW5G dpm++2m : https://vocaroo.com/19E6tT0itbIQ Both on : ultra_fast_old.

This did not happen prior to update in which we have "latent averaging mode" to select on the gui. Or at least I noticed this started to happen since then.

I have tested different voices and it always the same crackling sound. I haven't changed any of my voices either. Again not a big issue but I wonder if anyone else has noticed this?

152334H commented 1 year ago

latent averaging mode is actually a bit broken right now -- it discards up to 4.27s of each audio file -- so I'll set the default back to the tortoise original first

I'll also set the default to DDIM+10steps, because that's what the old ultra_fast impl used to use. dpm++2m continues to be experimental until I get the time to fix it

152334H commented 1 year ago

also I might add an extra voice fixer model to the pipeline

Angrod commented 1 year ago

I haven't tested from CLI but through GUI I'm also getting that crackling/mouse click sound only on the combined.wav file. They aren't there on the individual files.