tortoise-tts
tortoise-tts copied to clipboard
neonbjb
Voice / Speaker keeps changing mid sentence
No matter what I do, I don't even have any comma or period but the voice still changes completely to a different speaker mid sentence.
Something like "Hello there are you having a great day today on this most joyfull day" will have two speakers with completely different tone and pitch.
Tortoise is pretty cool, but it's basically useless when I have to sit through the generation only for it to generate something like that with two voices.
Is there any solution?
I had the same , but it didnt do it a few weeks ago so Im not sure what has changed.
How many audio clips are you using? I had the same issue when only using 3 or 4 clips. I'm using 62 10 second clips now and it sounds pretty good.
Yeah, some voices are better or worse than others. I have made a pull request to add more voices: https://github.com/neonbjb/tortoise-tts/pull/425
Each voice has dozens of audio clips.
Try them out yourself and tell me how you like them. They are from https://dillonbecker.itch.io/sdap
How many audio clips are you using? I had the same issue when only using 3 or 4 clips. I'm using 62 10 second clips now and it sounds pretty good.
I've tried as low as 2 clips and as high as 30 clips, no luck... voice changes every now and then.
I've had the same problem using one of the fine tuned voiced train_kennard.
I'm generally struggling to keep the voice consistent one sentence against another, it seems to change substantially. Any ways to reduce this?
I've had the same problem using one of the fine tuned voiced train_kennard.
I'm generally struggling to keep the voice consistent one sentence against another, it seems to change substantially. Any ways to reduce this?
i think this is just one of the limitations of the model/code as it seems to start iterating again with new metrics in a new sentence, reading from a text file seems better with stop tokens but still there is a change, or no change at all making it a monotone parrotting of words.