How do I get a similar result?

[bbecausereasonss]

I finally got a result that sounded like the input. There were 3 results with do_tts. 2 of them were way off one of them was on the mark. However, any time I try and redo it, (even with asking it to do 1) it always sounds different. How do I get consistency?

[dsunjka]

I was wondering about the same question - "how to keep the generated voice stable?", i.e. if I am using read.py for longer texts, I want all the sentences to sound "the same".

Setting the "seed" parameter didn't help here, I guess there are some more "random" things going on (?)

[bbecausereasonss]

Yea, I'm sometimes getting a completely different sounding voice with each sentence lol, like male, female, old, young

[embanot]

bump because I'm hoping someone can address this issue. Using read.py seems pointless if you're not able to get consistentcy with the voice quality/timbre

[neonbjb]

Hey there - unfortunately there's not a ton you can do. You can try increasing determinism by lowering the temperature, increasing top_p or increasing cond_free_k in the sampling parameters, but I'm not optimistic these things will work for this particular problem.