REDDIT 原始帖子
Live and local audio chat in 2026?
As bad as ChatGPT Advanced Voice is for counting to 100, I think it's pretty cool, and would like to see it on my desktop. We've got models like Gemma4 12B, Kokoro and the like. Are there any apps that tie everything together for an Advanced Voice like…
As bad as ChatGPT Advanced Voice is for counting to 100, I think it's pretty cool, and would like to see it on my desktop. We've got models like Gemma4 12B, Kokoro and the like. Are there any apps that tie everything together for an Advanced Voice like experience, or do I need to write one myself?
已收录讨论
You might like this: https://browserlab.missionsquad.ai/ You can run all of the models for this pipeline in your browser, STT -> llm/vlm -> TTS, all local. There’s a link to the SDK on there too.
This is fascinating! WebGPU inference on my phone! Thanks for the link! Looks like we're getting closer, but not quite the live, full duplex chat I was hoping for.
thanks! it's meant more as an example of what can be made - and the SDK makes it pretty easy to add any or all of that functionality to a web app. does full duplex exist? the closest thing might be using one of the gemma 4 onnx models, they can take audio and text as input (and images), and then just stream the output through one of the TTS models. but, yeah - probably not exactly what you're looking for, this only has small models.
I went the write one myself route, it's a STT/TTS engine. Low latency streaming over a websocket connection. I'm working on conversational turns right now. Basically you pipe your LLMs inputs and outputs through it. There seems to be a lot of people rolling their own in this space. Initially I wasn't planning on releasing mine but it's looking more and more like I will.