persian-asr-serving

Upload a file or record from your mic to see a transcript here.

Whole-file decode via POST /v1/transcribe — one request, one final transcript. No diarization yet (every token ships speaker: "1" as a placeholder, per docs/api-schema.md).

disconnected

Click Start, allow mic access, and speak. Settled text is solid; text still being decided is faded/italic. A dotted underline marks a word the graduated safety-net VAD flushed early (no natural pause found within 3–6s of speech) rather than a real end of turn.

ws://…/v1/transcribe-websocket — same protocol shape Soniox uses (docs/api-schema.md §3), driven here by poll-decode + a graduated safety-net VAD (docs/api-surface.md), not the native streaming export. Raw PCM only — this page resamples your mic to 16 kHz mono client-side via the Web Audio API.