How Speaker-Labeled Transcription Works
AnyTool transcribes and diarizes a recording entirely in the browser. A self-hosted Whisper ASR model (transformers.js on WebAssembly/WebGPU) converts speech to timestamped text, while local speaker diarization clusters the audio by voice to attribute each segment to a distinct speaker (Speaker 1, Speaker 2, …). The transcript is rendered with editable speaker labels and timings, and can be exported as a labeled transcript or as SRT that preserves speaker tags. All audio is decoded with the Web Audio API and processed on the device — nothing is uploaded and no CDN inference is used.
- Runs client-side — meeting audio is never uploaded
- Self-hosted Whisper ASR for the transcript
- Local speaker diarization groups speech by voice
- Editable, renameable speaker labels
- Export a labeled transcript and timed SRT
- No per-minute cap, account or CDN inference
