Insanely Fast Whisper: Blazing-Fast Audio Transcription CLI

Introduction

insanely-fast-whisper is a highly optimized command-line interface (CLI) for transcribing audio files using OpenAI's powerful Whisper models directly on your device. This project stands out for its incredible speed, capable of transcribing 150 minutes of audio in under 98 seconds when utilizing Whisper Large v3. It achieves this remarkable performance by integrating cutting-edge technologies such as Hugging Face Transformers, Optimum, and Flash Attention 2, making blazingly fast transcription a reality for developers and users alike.

Installation

To get started with insanely-fast-whisper, the recommended method is to use pipx. If you don't have pipx installed, you can install it first:

pip install pipx
# or on macOS
brew install pipx

Then, install insanely-fast-whisper:

pipx install insanely-fast-whisper

If you encounter issues with Python 3.11.XX parsing the version incorrectly, you can force the installation of the latest version:

pipx install insanely-fast-whisper --force --pip-args="--ignore-requires-python"

Examples

Once installed, transcribing audio files is straightforward.

Transcribe a file:

insanely-fast-whisper --file-name <filename or URL>

Note: For macOS users, remember to add --device-id mps flag.

Utilize Whisper-large-v3 with Flash Attention 2 for maximum speed:

insanely-fast-whisper --file-name <filename or URL> --flash True

Transcribe using Distil-Whisper models:

insanely-fast-whisper --model-name distil-whisper/large-v2 --file-name <filename or URL>

You can also run it without a full installation using pipx run:

pipx run insanely-fast-whisper --file-name <filename or URL>

For a full list of options and their defaults, run insanely-fast-whisper --help.

Why Use It

insanely-fast-whisper offers a compelling solution for anyone needing high-performance audio transcription. Its primary advantage is unparalleled speed, significantly reducing transcription times compared to standard Whisper implementations. The CLI is highly opinionated and optimized for NVIDIA GPUs and Apple Silicon (MPS), providing robust support across these platforms. It simplifies the process of leveraging advanced optimizations like Flash Attention 2 and various Whisper models, including Distil-Whisper, directly from your terminal. Whether you're a researcher, developer, or content creator, this tool provides an efficient and accessible way to convert speech to text with remarkable speed.

Links

• GitHub Repository