Welcome to LingoNaut! This repository contains a very simple python script for creating a custom speech-to-speech multilingual language learning assistant.
LingoNaut uses OpenAI's Whisper for speech-to-text, any Ollama model of your choice for the LLM, and the TTS package for text-to-speech.
- Make sure you have ffmpeg installed on your system with the location of the
bin/
folder added to your path. - Install Ollama on your system. Windows users will need to serve Ollama from WSL, but can then run client scripts from Powershell.
- Install conda or miniconda on your system.
- Navigate to this repo, and use
conda env create -f environment.yml
to install. - Use
conda activate lingonaut
to activate your environment. - Run
python create_lingonaut_ollama.py
to create the custom model. - Run
python lingonaut.py
to launch the session with the language assistant.
- There are no special options for which language to learn. All models used are fully multilingual, simply state your intention and let the assistant guide you.
- After running the Python script, you will see a message in the terminal that says "Awaiting user input..." when it is your move.
- To ask the assistant questions in English, hold down
Ctrl
and ask your question. On key release, your message will be passed to assistant. - When practicing another language, hold down
SHIFT
to use a larger version of the Whisper model which is more accurate in non-English transcription.
- To ask the assistant questions in English, hold down