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.