faster-whisper vs whisper.cpp vs WhisperX: which one should you use?

checked 2026-10-08 3 entries from the Tools Atlas

Short answer. All three are free and allow commercial use (MIT or BSD-2-Clause). Pick faster-whisper if you have a graphics card, whisper.cpp for other computers or a small program of your own, and add WhisperX when you need exact word timing or speaker labels. Use one of the first two, not both.

By Tori, last checked 2026-10-08. Each entry below was read against its own page or licence. The verdicts are written for someone who wants to use a tool in something they sell or publish; a skip is often about a licence or a heavy setup, not about quality.

At a glance

Name	Licence	Cost	Rules check	Atlas verdict
faster-whisper	MIT	Free software.	Allowed	Use it
whisper.cpp	MIT	Free software.	Allowed	Use it
WhisperX	BSD-2-Clause	Free software. Speaker labels need a free Hugging Face account token.	Allowed	Use it

One by one

faster-whisper Use it

Turns audio or video you own into a timestamped transcript on your own graphics card, several times faster than the original Whisper. MIT. With the Whisper large-v3 model it gives word timestamps for captions; the large-v3 model files are Apache-2.0.

Licence. MIT. Cost. Free software.

Source page, read 2026-10-08

whisper.cpp Use it

The same Whisper model in plain C and C++, with a built-in local web server. It runs on NVIDIA, AMD and Intel graphics cards and on a plain processor, so it is the better pick on a computer without an NVIDIA card, or for a small app. The large model needs about 3.9 GB. MIT.

Licence. MIT. Cost. Free software.

Source page, read 2026-10-08

WhisperX Use it

Adds exact timing for every word (for captions that follow the speech word by word) and speaker labels for two-person recordings, on top of faster-whisper. BSD-2-Clause. Speaker labels need a free Hugging Face account token.

Licence. BSD-2-Clause. Cost. Free software. Speaker labels need a free Hugging Face account token.

Source page, read 2026-10-08

Quick chooser

You have a recent graphics card and want transcripts or captions: faster-whisper.

Your computer has no NVIDIA card, or you are building a small program to hand to others: whisper.cpp.

You need captions timed to each word, or a podcast with two voices labelled: faster-whisper plus WhisperX.

What to know before you start

These run on your own computer, with no account at a service, so no rules of a platform apply. Use them on audio and video you own or have the right to use.

Run speech-to-text before translation or voice work if your graphics card is small, since each step wants the card to itself.

Questions

Is faster-whisper free for commercial use?
Yes. The code is MIT, and the Whisper large-v3 model files are Apache-2.0.

Should I install both faster-whisper and whisper.cpp?
No. They do the same job; pick one.

What do I need for speaker labels?
WhisperX, plus a free Hugging Face account token.

Want all 86 checked entries in one file?
Get the Tools Atlas, USD 9 See the free part first

PDF, EPUB, CSV and JSON in one zip, delivered by Gumroad right after payment. One payment. No income promise. Refunds: If this product is not useful to you, ask within 14 days of buying, from your Gumroad receipt or at hello@ticassociation.com, and you get all your money back.