WhisperX
Fast speech recognition with word timestamps and speaker diarization.
Tested on LinuxKnown vulnerabilities
Use it
Before you install: known vulnerabilities
pickfit found 9 known high or critical vulnerabilities in the packages a fresh install brings (OSV.dev, 5 Nov 2026). Weigh this before you install it.
pypi:torch 2.8.0PyTorch Vulnerable to Remote Code Execution via Untrusted Checkpoint Files (GHSA-63cw-57p8-fm3p, high, fixed in 2.10.0)pypi:transformers 4.57.6HuggingFace transformers vulnerable to remote code execution (GHSA-29pf-2h5f-8g72, high, fixed in 5.3.0)
See 7 more
pypi:torch 2.8.0— PYSEC-2025-203 (high), fixed in 2.9.0pypi:torch 2.8.0— PYSEC-2025-204 (high), fixed in 2.9.0pypi:torch 2.8.0— PYSEC-2026-139 (high), no fix yetpypi:transformers 4.57.6— GHSA-fgcw-684q-jj6r (high), fixed in 5.5.0pypi:transformers 4.57.6— GHSA-x9r9-c232-4q39 (high), no fix yetpypi:transformers 4.57.6— GHSA-xrqw-3rrv-vx5w (high), fixed in 5.10.0pypi:transformers 4.57.6— PYSEC-2025-217 (high), no fix yet
Linux x86_64
You need first
- System packages:
ffmpeg - A GPU is optional
- Keys or accounts: Hugging Face access token
Install whisperx via uv tool
uv tool install whisperxInstall static ffmpeg to ~/.local/bin
mkdir -p ~/.local/bin /tmp/ffdl && curl -L -o /tmp/ffdl/ffmpeg.tar.xz https://johnvansickle.com/ffmpeg/releases/ffmpeg-release-amd64-static.tar.xz 2>&1 && tar -xf /tmp/ffdl/ffmpeg.tar.xz -C /tmp/ffdl && cp /tmp/ffdl/ffmpeg-*-amd64-static/ffmpeg ~/.local/bin/ffmpeg && chmod +x ~/.local/bin/ffmpeg && ~/.local/bin/ffmpeg -version
Check that it works
whisperx --help 2>&1The output should contain “Show whisperx version information and exit”.
First thing to try
curl -L -o audio.wav https://www.voiptroubleshooter.com/open_speech/american/OSR_us_000_0010_8k.wav && ~/.local/bin/ffmpeg -hide_banner -y -i audio.wav -ar 16000 -ac 1 sample.wav && ls -lh sample.wavwhisperx sample.wav --model tiny --device cpu --compute_type int8 --language en --output_dir wxout 2>&1What it does
- Transcribe speech with word-level timestampsTested
- Align words using wav2vec2 modelsTested
- Label transcript with speaker IDsNot tested
How far pickfit got
Linux x86_64
- Installs
- Starts
- 2 of 3 features work
- Needs the network
Windows and macOS: the project says it runs there; pickfit hasn’t tried it.
Good for
- transcribe audio with speaker IDs
- generate subtitles with word timings
- transcribe non-English audio
An alternative to openAI's whisper.
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
Its main function works
- Replayed from scratch
- Needed the network
- Checked on 4 Nov 2026
- Version 3.8.6
Differences the test machine hit
- README: You may also need to install ffmpeg
In the test: downloaded johnvansickle static ffmpeg 7.0.2 tar.xz and copied ffmpeg binary to ~/.local/bin/ffmpeg on PATH - README: whisperx path/to/audio.wav (defaults) and --model large-v2 examples
In the test: ran whisperx sample.wav --model tiny --device cpu --compute_type int8 --language en instead of default small/GPU and large-v2 examples