Inside Scribe v3: speakers, timestamps and a review boundary
A recent speech-model release separates recorded and live audio, with documented API outputs a reviewer can trace back to the source.
Product Hunt’s 11 November 2026 daily board listed Scribe v3 by ElevenLabs fourth with 176 points when checked 12 November. This is a dated launch signal, not an independent accuracy test.
Documented components, not a complete production inventory.
- Scribe v3 / v3 Live
- Documented recorded-audio and live model routes
- Recorded-audio HTTP endpoint
- Documented multipart request and timed word/speaker response
- WebSocket live integration
- Official realtime transport documentation; no inference about backend stack
A model release from an existing provider
The Product Hunt board for 11 November lists Scribe v3 fourth with 176 points at this check. ElevenLabs published its official announcement on 10 November 2026. These are different dates and neither makes ElevenLabs a new company. FreeStack has not run Scribe or independently measured its accuracy.
Confirmed model and transport choices
The provider documents Scribe v3 for recorded audio and Scribe v3 Live for realtime audio, with a WebSocket integration for the live route. This confirms a split between batch and streaming interfaces. It does not establish the complete neural architecture, training dataset, cloud infrastructure or database stack; those details are not inferred here.
Confirmed output and request contract
The recorded-audio API uses POST /v1/speech-to-text with a selected model and uploaded media or a storage URL. Its example response contains word-level start/end values and speaker IDs. It also documents asynchronous webhook results. Those fields can support a review trail, but a structured response can still contain wrong words or labels. An ID is not a verified person.
Access and data handling remain separate
Current Creative pricing lists a $0 plan with 10,000 monthly credits and Speech to Text among its features; paid Starter adds a commercial licence. This check does not convert credits into promised Scribe v3 minutes or establish every model’s free eligibility. The API reference says logging defaults on and zero-retention mode is enterprise-only. Review the actual account, model, rights and retention requirements before sending a client recording.
A smaller achievable learning route
FreeStack editorial suggestion: use Audacity to replay an authorised clip and LibreOffice to log speaker turns, source times and proposed cuts. These are learning tools, not ElevenLabs’ production stack. The companion guide produces a paper edit, not a transcription engine. A service becomes worth paying for only when a named volume, language, review or retention requirement is demonstrated.
Sources & further reading
Follow the source for current details. Product Hunt is used for discovery and launch history; technical claims use provider documentation or the project repository.