Choose Whisper Web if…
- You need a transcript or SRT without setting up CUDA or Python
- Your computer has no NVIDIA GPU
- You want the file processed locally in the browser
Choose WhisperX if…
- You need automatic speaker labels
- You need word-level timestamps
- You process large batches fast on your own GPU
Whisper Web vs WhisperX at a glance
| Feature | Whisper Web | WhisperX |
|---|---|---|
| Setup | Whisper WebOpen a web page | WhisperXpip install; CUDA toolkit 12.8 for GPU; may need ffmpeg and Rust |
| Interface | Whisper WebBrowser editor with exports | WhisperXCommand line or Python |
| Speaker labels | Whisper WebNot provided | WhisperXWith pyannote; needs a Hugging Face token |
| Timestamps | Whisper WebSegment level | WhisperXWord level, with wav2vec2 alignment |
| Speed | Whisper WebDepends on your device | WhisperXAbout 70x realtime with large-v2 on a GPU, per its README |
| Cost | Whisper WebFree local mode; Unlimited US$20/mo | WhisperXFree under BSD-2-Clause; you supply the hardware |
WhisperX facts from its official pages, checked 8 Nov 2026. Confirm current plans on each site.
Setup
Whisper Web
Open the page and choose a file. There is no Hugging Face token to create and no model agreement to accept.
WhisperX
Install with pip, add a CUDA 12.8 toolkit for GPU use, and possibly ffmpeg and Rust. Speaker diarization needs a Hugging Face token and accepting the pyannote model terms.
Verdict
If you already run Python on a GPU, WhisperX is a short step. Otherwise, Whisper Web gets you a transcript in minutes.
Speakers and timing
Whisper Web
Whisper Web does not label speakers automatically; add names while you edit. Timestamps are per segment, which is what SRT and VTT need.
WhisperX
WhisperX aligns every word with wav2vec2 and labels speakers with pyannote.
Verdict
For interviews where you must know who said what, or for word-level karaoke subtitles, use WhisperX.
Where WhisperX is stronger
Automatic speaker diarization, word-level timestamps and very fast batched inference on a GPU. Whisper Web has none of these; it is for getting a Whisper transcript without an environment to maintain.
Frequently asked questions
Is there a WhisperX online version?
WhisperX itself is software you install. For Whisper transcription without installing anything, Whisper Web runs Whisper in your browser; it does not include WhisperX's speaker labels.
Does Whisper Web label speakers?
No. Add speaker names while you edit, or use WhisperX when automatic diarization is required.
Do I need a Hugging Face token?
Not for Whisper Web. WhisperX needs one for speaker diarization with pyannote.
Which is faster?
On a capable GPU, WhisperX; its README cites about 70x realtime with large-v2. Whisper Web's speed depends on your browser and device.