robit

Proof-of-concept of a conversational robit using GPT, WhisperAI, and Mimic 3. Inspired by https://www.youtube.com/watch?v=bO-DWWFolPw.

Running

1. Install requirements:
   • python -m pip install -r requirements.txt
     • You may have to install some additional libraries depending on your system. For example, in Ubuntu, you may have to run sudo apt install portaudio19-dev.
2. Get yourself a running transcription service. I've been using whisper-jax with good success.
3. If you want to use Mimic 3 instead of Open AI for TTS:
   • Spin up Mimic 3, for example using docker: https://mycroft-ai.gitbook.io/docs/mycroft-technologies/mimic-tts/mimic-3#docker-image
   • Use Custom_TTSHelper instead of OpenAI_TTSHelper in AIHelper's __init__.py.
   • Change the format argument in AudioHelper.say() to wav (Mimic 3 outputs a wav, OpenAI an MP3).
4. Put your settings in the env (or stick them in some other way, I'm not your mom).
   • This project uses dotenv, so all you have to do is create a .env file in the root of the repo.
   • Supported settings:
     • OPENAI_API_KEY - API key for OpenAI. Required (chat responses use GPT, and TTS uses OpenAI's TTS model by default)
     • ROBIT_TRANSCRIPTION_ENDPOINT - Endpoint for a transcription service. Defaults to http://localhost:4444/transcribe.
     • ROBIT_TTS_ENDPOINT - Endpoint for a TTS service. For use with Custom_TTSHelper. Defaults to http://localhost:59125/api/tts (the default Mimic 3 URL).
     • ROBIT_LOG_LEVEL - Specify output log level. Recommend setting this to DEBUG especially for now.
5. Run robit.py.
   • python robit.py (or some other way. I don't know your life. Calm down.)