Make two AIs speak to each other on a given topic with a voice.

Built using Ollama, Llama3, Coqui-TTS.

Both the text generation and audio generation is done locally.

The voice server is written in Python. The text bubble server is written in Node.js.

Getting started

First install dependencies.

pyenv install 3.11
pyenv virtualenv 3.11 aichatsai
pyenv activate aichatsai
pip3 install -r ./requirements.txt
npm install

Then, run the following commands in three separate terminals. (I'm working to combine them into one command.)

pyenv shell aichatsai
python3 ./src/voice.py

pyenv shell aichatsai
python3 ./src/voice_player.py

npm run start -- --icebreaker 'How to be a better software developer?'

Example

Unmute the video to listen.

Note that there is a slight jitter in the audio. I do believe the jitter is due to excessive CPU workload - since I'm running both audio and text inferences on the CPU locally.

ai-chats-ai-new-screenrecord.mp4