VoiceGPT is a voice assistant that leverages the powerful ChatGPT chatbot to answer your questions. You speak the requests, and VoiceGPT responds with realistic, synthesized speech.
I chose a Raspberry Pi 4 single board computer to host the project, because it runs Linux and provides a lot of versatility. A custom Python script collects audio of a speaker's voice using a USB microphone. The Google Cloud Speech-to-Text API is then used to convert that audio file into text. The text is then queried against ChatGPT using an unofficial API that returns a text string of ChatGPT's response. That response is then processed by Google Cloud's Text-to-Speech API to turn it into realistic, synthetic speech that the Raspberry Pi can play through a speaker.
The concept of a voice assistant is well established (e.g. Google Home, Amazon Alexa), but this proof of concept shows how a voice assistant can use ChatGPT, which, in my opinion, provides a far better experience than anything currently on the market.
In the future, I may add a keyword spotting algorithm to the project so that it can always run in the background, waiting for a keyword (e.g. "Hey, ChatGPT") to wake up. Before I have the chance to do much of anything else, there will probably be a commercial product including ChatGPT on the backend — then I'll just buy that because it will be smaller and better. :)
- 1 x Raspberry Pi 4
- 1 x USB microphone (I use a webcam with a built-in microphone)
- 1 x Speaker
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent