3 ms·
Cool. I built something similar for myself. Though it uses the GPT-3 API rather than ChatGPT: Whisper.cpp for text input. Google WaveNet Voice for text output.
by leobg 4y ago
Cool. I built something similar for myself. Though it uses the GPT-3 API rather than ChatGPT:
Whisper.cpp for text input. Google WaveNet Voice for text output. One button for the user to start and stop speaking, like Siri. Allows me and my daughter to literally talk to GPT-3 and have real conversations with it.
(Though I’d never let her do that without supervision. Also, she has learned very quickly that she needs to take everything it says with a grain of salt, and that it’s important to fact-check its answers.)
I’d be happy to show it, but Whisper is quite CPU intensive. I don’t know how to host it so it can handle any meaningful number of concurrent users without breaking the bank. If anyone has suggestions or wants to help, let me know.
- louison11 4y agoUse a cloud service for speech to text, like Google’s API, and you should be able to handle a pretty good number of users without breaking the bank. (I believe they have a free tier, then pretty reasonable pricing. You’d just have to set up a rate limiter on your server to make sure nobody’s abusing it).
- jsharf 4y agoTheoretically, one could compile whisper.cpp to run in the browser using emscripten, maybe made faster with webgl... I think this would be quite a heavy page load time for a website, but if the model file gets cached, and the user has a decent CPU/GPU, it... could work?
- leobg 4y agoThat is a cool idea! And I just realized the Whisper.cpp guy has already built something like this. Whisper WASM. [1] Unfortunately, on my iPhone this runs so slow that I might just as well send my mic recordings to a typist. And mobile is important here. Being able to quickly pull this up on your phone really is part of the fun. [1] https://whisper.ggerganov.com/ https://whisper.ggerganov.com/