speechless

UV Installation Instructions

To install dependencies and manage the project, we use uv, a fast Python package manager and resolver. Follow the steps below to set up your environment.

Step 1: Install uv

You can install uv via pip:

pip install uv

Or with pipx:

pipx install uv

Verify the installation:

uv --version

Step 2 · Create (and activate) the virtual environment

Use uv to manage your environment:

uv venv source .venv/bin/activate

(If you prefer another tool like venv/virtualenv, activate it before continuing.)

Step 3 · Install dependencies

Install all runtime, dev, and extras in one step:

uv sync --all-extras --dev

This also creates or updates uv.lock automatically.

Step 4 · Run the project or test suite

If not already activated:

source .venv/bin/activate

Then run the tests:

uv run pytest

Step 5 · Run pre-commit hooks

Ensure your code stays clean:

uv run pre-commit run --all-files

Quick commands you might love:

Step 6: Convert the model to ONNX format

To convert the model to ONNX format, run:

python export_to_onnx.py --checkpoint /path/to/checkpoint --onnx_model /path/to/onnx_model

Step 7: Add OPENAI_API_KEY and/or Set Up WHISPER_CPP_MODEL

The whisper_1 model requires an OpenAI subscription. As an alternative, you can use whisper.cpp.

To download a supported model:

# Linux
docker run -it --rm -v ./data/models:/models ghcr.io/ggerganov/whisper.cpp:main "./models/download-ggml-model.sh small /models"

# Windows (PowerShell)
docker run -it --rm -v "$(pwd -W)/models":/models ghcr.io/ggerganov/whisper.cpp:main "./models/download-ggml-model.sh small /models"

Once WHISPER_CPP_MODEL is set, inference is handled locally:

ffmpeg -i data/temp_results/uploaded_audio.mp3 -ar 16000 -ac 1 -c:a pcm_s16le data/audio/output.wav

Run whisper.cpp:

# Linux
docker run -it --rm -v ./data/models:/models -v ./data/audio:/audios ghcr.io/ggerganov/whisper.cpp:main "./build/bin/whisper-cli -m /models/ggml-small.bin -f /audios/output.wav -ml 16 -oj -l en"

# Windows
docker run -it --rm -v "$(pwd -W)/data/models":/models -v "$(pwd -W)/data":/audios ghcr.io/ggerganov/whisper.cpp:main "./build/bin/whisper-cli -m /models/ggml-small.bin -f /audios/output.wav -ml 16 -oj -l en"