Repolode
Submit
SYSTRAN

SYSTRAN/faster-whisper

A faster reimplementation of OpenAI's Whisper speech-to-text model on the CTranslate2 inference engine, with lower memory use and 8-bit quantization.

Open-source alternative to:Otter.aiRev

View on GitHub
Stars
25,987
Forks
2,152
Open issues
20
Last commit
4 days ago

What is it?

faster-whisper is a Python reimplementation of OpenAI's Whisper speech recognition model that runs on CTranslate2, a fast inference engine for Transformer models. It is aimed at developers who want to transcribe audio with Whisper while spending less time and memory on inference, and it supports 8-bit quantization on both CPU and GPU.

The library exposes a WhisperModel class that loads models by size name, such as large-v3 or distil-large-v3, and downloads converted weights from the Hugging Face Hub. Transcription returns a generator of segments with timestamps, and options include word-level timestamps, Silero VAD-based silence filtering, and a BatchedInferencePipeline for batched GPU or CPU processing. Audio is decoded through PyAV, so FFmpeg does not need to be installed separately.

A conversion script turns Transformers-compatible Whisper models, including fine-tuned ones, into the CTranslate2 format. GPU use requires NVIDIA cuBLAS and cuDNN libraries, with version notes for CUDA 11 and 12.

Key features

  • 01Runs Whisper models on CTranslate2 with lower memory use than the original implementation.
  • 02Supports FP16 and INT8 quantization on both GPU and CPU.
  • 03Provides word-level timestamps and segment timestamps.
  • 04Includes Silero VAD filtering to skip silent parts of audio.
  • 05Offers a batched inference pipeline that is a drop-in replacement for transcribe.
  • 06Converts Transformers-format Whisper models, including fine-tuned ones, to CTranslate2.

Use cases

  • Transcribing meeting recordings or podcasts on a local GPU or CPU server.
  • Adding speech-to-text to a Python application or backend service.
  • Generating subtitles with word-level timing for video content.
  • Building a real-time or near-real-time transcription pipeline on top of Whisper.

Quick start

  1. 1

    Install the package from PyPI

    pip install faster-whisper
  2. 2

    Transcribe an audio file with the Python API (see the Usage section for the full code)

Things to know

  • GPU use requires separately installed NVIDIA cuBLAS for CUDA 12 and cuDNN 9 libraries.
  • The latest ctranslate2 only supports CUDA 12 and cuDNN 9, so older CUDA setups need a downgraded ctranslate2 version.

Topics

#speech-to-text#whisper#speech-recognition#ctranslate2#quantization#transcription#inference#deep-learning#openai#transformer

FAQ

What is faster-whisper?
A faster reimplementation of OpenAI's Whisper speech-to-text model on the CTranslate2 inference engine, with lower memory use and 8-bit quantization. faster-whisper is a Python reimplementation of OpenAI's Whisper speech recognition model that runs on CTranslate2, a fast inference engine for Transformer models. It is aimed at developers who want to transcribe audio with Whisper while spending less time and memory on inference, and it supports 8-bit quantization on both CPU and GPU.
Is faster-whisper free to use? What license does it have?
faster-whisper is open source under the MIT license. Check the license terms before using it in commercial products.
Is faster-whisper still maintained?
Actively maintained. The last commit to faster-whisper was on Oct 16, 2026.
What can I use faster-whisper for?
Transcribing meeting recordings or podcasts on a local GPU or CPU server. Adding speech-to-text to a Python application or backend service. Generating subtitles with word-level timing for video content. Building a real-time or near-real-time transcription pipeline on top of Whisper.
How do I install faster-whisper?
Install the package from PyPI: pip install faster-whisper Transcribe an audio file with the Python API (see the Usage section for the full code)
Where can I get faster-whisper?
The source code and installation instructions are on GitHub: https://github.com/SYSTRAN/faster-whisper

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects