ElevenLabs Scribe speech-to-text

Use this skill when a media-production task needs ElevenLabs Speech to Text (Scribe) for transcripts, caption timing, speaker-labeled interview logs, podcast edit prep, accessibility subtitles, localization handoff, call/interview analysis, or transcript cleanup. Do not use it as a text-to-speech, dubbing, voice cloning, or translation skill; Scribe produces text, timing, and metadata from existing audio/video. Use dubbing or TTS skills only after the transcript is verified and the user has rights to transform the source.

Volatile facts below were verified from official ElevenLabs documentation on 2026-07-10. If building against the live API, re-check model names, limits, pricing, residency endpoints, and retention options before irreversible production work.

Source-grounded facts

Current models and fit

• Use scribe_v2 for batch/file transcription. ElevenLabs documents Scribe v2 as supporting 90+ languages, word-level timestamps, speaker diarization up to 32 speakers, dynamic audio tagging, keyterm prompting, entity detection, and smart language detection.
• Use scribe_v2_realtime only for live streaming use cases such as agents, live meetings, or interactive monitoring. It is a WebSocket STT service, not the normal path for post-production captions from a finished file.
• Treat scribe_v1 as retired/unsafe for new work. The API schema still listed scribe_v1 as an allowed value when checked, but the official changelog announced Scribe v1 removal on 2026-07-09. Prefer scribe_v2 and flag any Scribe v1 dependency as a migration issue.

Inputs, outputs, and request shape

Batch endpoint: POST https://api.elevenlabs.io/v1/speech-to-text with multipart form data.

Inputs: