Official documentation

faster-whisper: README (GitHub, SYSTRAN)

SYSTRAN, GitHub (SYSTRAN/faster-whisper), 2026

Read the original

Read on October 17, 2026

What it is: The README of faster-whisper, a reimplementation of OpenAI's Whisper on the CTranslate2 inference engine. It gives the install commands, the GPU requirements, the default voice activity filter and two benchmark tables comparing openai/whisper, whisper.cpp and faster-whisper.
How much weight it carries: Written and maintained by the project's own developers, so the speed claims are self-reported. The benchmark hardware, versions and audio are stated in the README, which makes the comparison checkable. Results will change with new releases.

Facts taken from this source

The faster-whisper benchmark measured how long each implementation took to transcribe a 13-minute recording.
“to transcribe [**13 minutes**](https://www.youtube.com/watch?v=0u7tTptBo9I) of audio” (Benchmark)
Running faster-whisper on a GPU requires NVIDIA libraries to be installed first.
“GPU execution requires the following NVIDIA libraries to be installed:” (GPU)
In the same CPU benchmark, faster-whisper with int8 precision took 1 minute 42 seconds (102 seconds) with the small model.
“| faster-whisper | int8 | 5 | 1m42s | 1477MB |” (Small model on CPU table)
In the faster-whisper README's GPU benchmark, faster-whisper in int8 precision used 2926 MB of GPU memory with the large-v2 model.
“| faster-whisper | int8 | 5 | 59s | 2926MB |” (Large-v2 model on GPU (benchmark table))
On the same GPU, faster-whisper with int8 precision transcribed the recording with the large-v2 model in 59 seconds.
“| faster-whisper | int8 | 5 | 59s | 2926MB |” (Large-v2 model on GPU table)
Unlike openai-whisper, faster-whisper does not need FFmpeg installed on the system, because it decodes audio with PyAV, which bundles the FFmpeg libraries.
“Unlike openai-whisper, FFmpeg does **not** need to be installed on the system.” (Requirements)
In the faster-whisper CPU benchmark, openai/whisper took 6 minutes 58 seconds (418 seconds) with the small model in fp32 precision.
“| openai/whisper | fp32 | 5 | 6m58s | 2335MB |” (Small model on CPU table)
In the faster-whisper GPU benchmark, openai/whisper took 2 minutes 23 seconds (143 seconds) with the large-v2 model in fp16 precision.
“| openai/whisper | fp16 | 5 | 2m23s | 4708MB |” (Large-v2 model on GPU table)
faster-whisper is a reimplementation of OpenAI's Whisper model that runs on the CTranslate2 inference engine.
“**faster-whisper** is a reimplementation of OpenAI's Whisper model using [CTranslate2](https://github.com/OpenNMT/CTranslate2/), which is a fast inference engine for Transformer models.” (Introduction)
faster-whisper is up to 4 times faster than openai/whisper for the same accuracy, while using less memory.
“This implementation is up to 4 times faster than ... for the same accuracy while using less memory.” (Opening paragraph)
The default voice activity filter in faster-whisper removes only silences longer than 2 seconds.
“The default behavior is conservative and only removes silence longer than 2 seconds.” (VAD filter)
In the same CPU benchmark, whisper.cpp took 2 minutes 5 seconds (125 seconds) with the small model in fp32 precision.
“| whisper.cpp | fp32 | 5 | 2m05s | 1049MB |” (Small model on CPU table)

faster-whisper: README (GitHub, SYSTRAN)

Facts taken from this source

Pages that rest on this source

faster-whisper: README (GitHub, SYSTRAN)

Facts taken from this source

Pages that rest on this source