peper0/live_transcribe

A Python package that provides live, real-time transcription of audio based on OpenAI's Whisper API.

Live Transcribe

Live Transcribe is a Python package that provides live, real-time transcription based on OpenAI's Whisper. Works offline.

Currently, Live Transcribe supports only PulseAudio as an audio backend.

Installation

1. optional but highly recommended for low latency Refer to the Pytorch documentation guide to install Pytorch with CUDA support.
2. pip install live-transcribe

Usage

Just run:

python -m live_transcribe

or

live_transcribe

If you want to transcribe from another audio device, than the default, use the --device option, e.g.:

live_transcribe --list-devices
live_transcribe --device "alsa_input.usb-046d_HD_Pro_Webcam_C920_8C0B5B0F-02.analog-stereo

On the first usage, the OpenAI's Whisper model will be downloaded and cached.

See live_transcribe --help for options.

Dependencies

Live Transcribe has the following dependencies:

Python 3.8 or higher
OpenAI-Whisper
PulseAudio