← sources

Generating unique (realistic) voices for TTS?

source: reddit (reddit:LocalLLaMA)published 2026-08-27 17:55ingested 2026-08-28 11:38

retained text

Eleven Labs has a "Voice Design" feature that lets you create a custom voice for TTS without voice-cloning a real person. But that's the opposite of local.

Are there any local tools that provide anything along that line? More customization is better but there's a broad range between static pretrained voices and celebrity deepfakes that would work for me.

It doesn't even actually have to do the TTS natively. I'd be totally fine cloning a fake voice in as long as the source sample is high enough quality for good results.

(The fallback is probably to sign up for Eleven Labs long enough to create a bunch of custom voices and have them all say "Hi, my name is Werner Brandes. My voice is my passport. Verify me." And then clone from that. But I do want to stay local whenever I can.)

Thanks for any suggestions!

document fetched 2026-08-28 11:27 · canonical https://reddit.com/r/LocalLLaMA/comments/1w017py/generating_unique_realistic_voices_for_tts