@jaybrunet@podcastindex.social
Post #2382474
2026-04-05 05:01 UTC
FWIW the default voices are retarded, try the cloning feature.
The separate "voice design" model is fun, BUT it won't repeat the same voice twice. If you really like a voice you 'designed' -- I suppose you could clone it, and then it's reusable -- if that makes sense.
Also important, the "instruct" feature does not work at all in the 0.6b model. It's supposed to work with the 1.7b model, but I found it doesn't make much difference with my cloned voice...
Replies (1)
-
@jaybrunet@podcastindex.social 2026-04-05 05:07
To expand on that, I have not tried the official Qwen3 TTS UI yet, partly bc @adam@podcastindex.social said it was a pain to install, so I have nothing to compare with. I suspect maybe Gabriele is doing something wrong with the 1.7b model. It does work, but like I said, the "instruct" option doesn't make any meaningful difference, even if you change the temperature. Is it just me? Try for yourself. Regardless, I'm really impressed with the speed & ease of use of the 0.6b model, especially if you prefer a CLI.