faster-whisper runs OpenAI's Whisper speech-to-text models through the CTranslate2 inference engine, so audio is transcribed faster and with less memory.
In plain words
Whisper turns speech into text, but the original code can be slow. faster-whisper runs the same Whisper models through CTranslate2, an engine built to run transformer models quickly. Its authors report it is up to four times faster at the same accuracy, and 8-bit quantization can speed it up further on CPUs and GPUs. It can also skip silent stretches using a voice activity detector and give timestamps for each word.
A quick example
Imagine an hour-long podcast recording. A short Python script loads a Whisper model with faster-whisper, asks it to transcribe the file with silence filtering switched on, and prints each segment of text with its start and end times, ready to turn into subtitles.
Where you'll meet it
- Turning audio files into text with word timestamps.
- Speech-to-text on ordinary CPUs or GPUs.
- Tools built on it, such as the whisper-ctranslate2 command-line client.
Related
Sources
- DocsSYSTRAN/faster-whisper, SYSTRAN (GitHub) · read 29 Sept 2026
- DocsOpenNMT/CTranslate2, OpenNMT (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.