Overview

This repository provides an asynchronous audio transcription utility using OpenAI’s Whisper model. The script reads audio files, splits them into 10-minute chunks (to keep each chunk under ~25 MB), and concurrently transcribes each chunk. The final transcription is then compiled and written to disk, along with an optional segments file containing time-stamped text segments.

Features

Uses asyncio to handle file I/O and concurrent chunk transcriptions for efficiency.

Configurable language

You can optionally pass a two-letter language code (ISO 3166-1 alpha-2) to the script (e.g., en, fr, es). The default is English (en).

Chunk-based processing

The script splits the audio into 10-minute chunks (configurable via CHUNK_DURATION_MS), making it easier to handle large audio files and keep them under the size limits for the Whisper API.

Transcribed segments with timestamps

The script can produce a separate segments file (_segments.txt), displaying each recognized text segment with start and end timestamps.

Prerequisites

• Python 3.9.10 (required for pydub compatibility)
• uv package manager
• OpenAI API key
• Audio file in supported format (mp3, wav, etc.)

Installation

1. Clone the repository and cd into the root directory:
2. Install uv (if not already installed). Read more here: uv

curl -LsSf https://astral.sh/uv/install.sh | sh

3. Create and activate virtual environment:

uv venv --python=python3.9.10
source .venv/bin/activate

4. Install required packages with uv:

uv sync

Usage

1. Create .env file with your OpenAI API key:

OPENAI_API_KEY=your_key_here

2. Run the script (with optional language code):

python whisper.py /path/to/audio.mp3 [language_code]

Important notes

• Python version 3.9.10 is required for pydub compatibility
• Large audio files will be processed in chunks
• Make sure your audio file is in a supported format (mp3, mp4, mpeg, mpga, m4a, wav, and webm.)
• See OpenAI's Whisper API documentation for more information