Qwen 3 TTS - New TTS bot, free demo online (with use limit), appearently easy to locally run - With love from Zhongguo

[BigMan13]

Github - Demo - Download

Showcase video: Youtube / Preservetube

Appearently a new voice synthesizer is better than usual. The guy claims its easy to set up, though it seems "easy" as in using command lines. Unclear if voice inserting is something locally runnable or to what extent. The online demo has about 3-4 uses per day for non-premium users, already hit the limit before I could include anything funny.

 

[Barry Convex]

I used Gemini to slap together a program to use Qwen3-TTS and I've been pretty impressed with the results so far. Having a good reference audio clip around 10-15 seconds that's clear is key for voice cloning. You can also add delivery instructions, but it's so-so on how much of that instruction is really heard in the output.

On an M4 Macbook Air, it's pretty darn fast, sometimes taking as low as 2 seconds of generation for 1 second of output. Definitely a lot of fun to play around with, thanks China 8)

 

[bearycool]

I like Qwen3 as a model, especially with making react components for websites. I need to lookm more into the gemini side of making a program as well, speaking of their CLI.


Oddly, when I was working as a contractor, when I used my PMLL (memory architecture shit in C), it was able to exit out of the sandbox and maintain past context memory without causing tjhe react component to break, nor was context rot a problem. Developer Mode, but roided and better than endless MCP tools like with claude code.

I need to look more into this later, thanks for making this thread!

 

[Overly Serious]

Impressive. I've been wanting to try this out since you posted the thread but today was the first opportunity I got. Running it locally and am quite impressed. And it's certainly small!

Note for people trying this with AMD cards, it will run with ROCm and seems fairly performant. You'll need to install a version of Torch with ROCm (I used version 7.1) and I can't compare it with Nvidia as I don't have a card to do so with (nor would I bother swapping it over) and I have a hunch it's running slower but it seems to give equivalent results to their online demo and is certainly fast enough to be useful.