faster-whisper
A reimplementation of OpenAI's Whisper speech-to-text models on top of CTranslate2, giving the same transcription quality as the reference models at several times the speed and a fraction of the memory, with int8 and other quantisations built in. It is a library rather than an app, so it is what you reach for when building transcription into something, rather than transcribing the occasional file by hand.
Strengths
- Up to about 4x faster than the reference Whisper at the same accuracy, using less memory
- Supports 8-bit quantisation on both CPU and GPU
- Compatible with the original and fine-tuned Whisper models
Weaknesses
- The GPU path needs specific NVIDIA libraries (CUDA, cuBLAS, cuDNN)
- A Python library rather than an app, so it needs setup
- Output is a generator, which trips up newcomers
At a glance
- Licence
- MIT
- Pricing
- Free
- Platforms
- macOS, Windows, Linux, Docker
- GPU required
- No
Links
Works with
Related guides
See also
Glossary
Entry last verified 18 August 2026.