PARRoT: Precise Audio Recognition and Recap over Transcription

A tool for writing a recap mail or report from a video recording of a call.

Installation

PARRoT can be installed from pyPI by simply doing:

python -m pip install parrot1

with the following extras:

faster-whisper -> to install faster-whisper dependencies
llama-cpp -> to install llama-cpp dependencies
os-models -> to install both the previous ones
docx -> to use external transcription
all -> install everything

For instance:

python -m pip install parrot1[os-models]

NOTE: PARRoT requires FFmpeg available and visible in the system!

Usage

You can find all the information needed for using the tool by doing

parrot --help

At the moment, there are two commands:

parrot mail <video_path> and parrot report <video_path>

Options:

--transcript (-t) <PATH>: path to the transcript with speakers (Optional).

--output-filepath (-o) <PATH>: path to the final result (Optional, by default it creates the file in the current working directory).

--faster-whisper (-fw): If set, it uses faster whisper instead of hosted OpenAI Whisper.

--llama-cpp (-lc): If set, it uses llama-cpp instead of the hosted OpenAI GPT model.

Contributing

First of all make sure you have installed pre-commit and can use it.

pip install pre-commit

and

pre-commit install

You can't push your code directly in the contrib branch. You should open a branch (feat|fix)/[/] and then open a PR with reviewers.

Please try sticking to the code style or, for any eventual suggestion for libraries/patterns, discuss it before in a issue.