faster-whisper
Faster Whisper transcription with CTranslate2
What is faster-whisper?
Faster Whisper transcription with CTranslate2
Overview of faster-whisper
What it is
faster-whisper is a reimplementation of OpenAI's Whisper model using CTranslate2, a fast inference engine for Transformer models. It is developed by SYSTRAN and aims to provide faster transcription with lower memory usage while maintaining the same accuracy as the original Whisper.
Key features
- Up to 4 times faster than openai/whisper for the same accuracy
- Lower memory usage compared to the original implementation
- Supports 8-bit quantization on both CPU and GPU for further efficiency
- Batched inference pipeline for processing multiple audio segments in parallel
- Voice Activity Detection (VAD) filter using Silero VAD to remove non-speech parts
- Word-level timestamps for precise alignment
- Compatible with Distil-Whisper checkpoints for faster inference
Use cases
- Transcribing audio files with high speed and low memory footprint
- Batch transcription of long audio recordings for efficiency
- Using Distil-Whisper models for even faster inference on supported hardware
- Obtaining word-level timestamps for subtitle generation or alignment
- Filtering out silence or non-speech segments with VAD before transcription
Getting started
Install the module from PyPI:bashpip install faster-whisperPython 3.9 or greater is required. GPU execution requires NVIDIA cuBLAS for CUDA 12 and cuDNN 9 for CUDA 12. FFmpeg is not needed because audio decoding uses PyAV which bundles FFmpeg libraries.
License and pricing
This project is licensed under the MIT License and is open source (free to use).
Based on the project's public README and repository metadata.