25.3k stars · MIT · v1.2.1 (2025-10-31)
Whisper transcription up to 4x faster with less memory, and it runs on plain CPU.
▶Repo detailsthe review · specs · pros & cons · install
What it is
faster-whisper is a reimplementation of OpenAI's Whisper model using CTranslate2, a fast engine for running this kind of model. It produces the same text as Whisper, with word-level timings, and runs on a processor alone if you have no graphics card.What it is good for. Anyone who records audio or video regularly and wants the text. Edition 2 recommended whisperX and Edition 13 recommended whisper.cpp. This is the third answer to the same question, and it is worth saying plainly which is which. whisperX is still the best at telling speakers apart. whisper.cpp is the easiest to install on a Mac and needs nothing from Python. faster-whisper is the one to use from Python code, and it is the fastest of the three on a normal processor. For a channel published in two languages, the transcript is the thing three other jobs depend on.
- About four times faster than the original, with lower memory use.
- It runs on an ordinary processor, so no graphics card is needed.
- It is a Python library, so it drops straight into a script that then makes the subtitle file.
- The latest release is 1.2.1 from 31 October 2025, and the last code landed 19 November 2025. That is within our eighteen-month line, but it is quieter than the alternatives.
- The first run downloads a model file of several hundred megabytes. Do that once, on a good connection.
- Accuracy on Russian is good but not perfect. Plan to read the transcript before it becomes a subtitle.
cd ~/youtube python3 -m venv venv source venv/bin/activate pip install faster-whisper