AV Tools

A collection of CLI tools for audio and video processing.

• Audio transcription and diarization
• Transcript formatting
• Video to audio conversion
• YouTube video downloader

• Python 3.11
• FFmpeg v7 full build and add the bin folder to the PATH environment variable.

To use diarization feature, you must have a Hugging Face account and follow these steps:

1. Accept pyannote/segmentation-3.0 user conditions
2. Accept pyannote/speaker-diarization-3.1 user conditions
3. Create access token at hf.co/settings/tokens.

avtools can be installed using pipx. If you don't have pipx installed, you can install it using pip (pip install pipx and python -m pipx ensurepath) or brew (brew install pipx and pipx ensurepath).

To install the package using pipx, run the following command:

pipx install git+https://github.com/jorgeandrespadilla/avtools.git

To upgrade the package, run the following command:

pipx upgrade avtools

avtools <command> [options]

For more information on the available commands, use the --help argument:

avtools --help

avtools transcribe -i <path_to_audio_file>.mp3 -o <path_to_output_file>.json

To use diarization feature, add the --hf-token argument with the access token. We do not recommended to use this feature for large audio files.

avtools video-audio -i <path_to_input_video_file>.mp4 -o <path_to_output_audio_file>.mp3

Only available for transcripts generated in JSON format.

To convert a JSON transcript to a subtitle file or plain text file, use the following command:

avtools format -i <path_to_input_json_file>.json -o <path_to_output_file>.srt

Supported output formats:

• srt
• txt
• vtt

avtools youtube-download -u <youtube_video_url> -o <path_to_output_file>.mp4

To download the video transcript, add the --transcript argument with the language code (e.g. en for English).

avtools youtube-download -u <youtube_video_url> -o <path_to_output_file>.mp4 --transcript=<language_code>

How to use Flash-Attention with avtools transcribe command?

Install it via pipx runpip avtools install flash-attn --no-build-isolation.

We only recommend using Flash-Attention if your GPU supports it.

See the CONTRIBUTING.md file for more information on how to contribute to this project.

The following resources may be helpful when solving issues related to PyTorch package installation with Poetry:

• python-poetry/poetry#7685
• Vaibhavs10/insanely-fast-whisper#183
• python-poetry/poetry#2415

Due to the way PyTorch is built, the source URLs have to be hard-coded in the pyproject.toml file to avoid installation issues (to support new Python versions, we should add more URLs to the torch packages). This is a workaround to avoid issues when working with private repositories.

This project is licensed under the MIT License - see the LICENSE file for details.