Speaker Diarization Online

Upload audio and see who spoke when. Free speaker diarization with timestamps, powered by the open-source pyannote segmentation model and Whisper — running in your browser.

Speaker diarization tool

Tool: open-source Hugging Face Space Xenova/whisper-speaker-diarization (models: pyannote-segmentation-3.0, MIT; Whisper base, Apache-2.0). Needs a modern browser; WebGPU recommended. If the frame does not load, open it on Hugging Face.

What is speaker diarization?

Speaker diarization answers the question “who spoke when?”. It splits a recording into segments and assigns each one a speaker label (Speaker 1, Speaker 2, …) with start and end timestamps. It is used for meeting notes, interviews, podcasts, call analytics and subtitle editing — anywhere you need to know which person said what.

This tool uses pyannote's segmentation-3.0 model to detect speakers and Whisper to transcribe each segment, so you get a speaker-labeled transcript.

How to use

  1. Load the model — click the load button in the tool above (first time only; models are cached).
  2. Add audio — upload an audio/video file or pick the example clip.
  3. Run — click run and read the timeline: each segment shows the speaker, timestamps and text.

FAQ

Is this speaker diarization tool free?
Yes. It is free with no sign-up. The tool runs in your browser (WebGPU/WASM via Transformers.js), so there are no server costs and no usage limits from us. The first run downloads the models (a few hundred MB) and caches them.
How accurate is it?
It uses pyannote segmentation-3.0, a widely used open-source speaker segmentation model. It works best on clean audio with 2–3 speakers who do not talk over each other. Heavy overlap, background noise, phone-quality audio, or many similar voices reduce accuracy. Check the timestamps before relying on them.
Which languages does it support?
Speaker diarization itself is language-independent: it separates voices, not words. The transcript shown next to each speaker comes from Whisper (base), which supports about 100 languages, with best results in English and other high-resource languages.
What is the difference between diarization and transcription?
Transcription turns speech into text (what was said). Diarization labels who spoke when (Speaker 1 from 0:00–0:12, Speaker 2 from 0:12–0:30). Combining them gives a speaker-labeled transcript.
Is my audio uploaded?
No. The embedded tool processes audio locally in your browser. Model files download from Hugging Face; your audio stays on your device.

Related tools

Only need text, fully on-device? Whistle Transcription Online — tiny speech-to-text that runs in your browser.