Intel / article

How Scribe V2 Realtime Works

Source: elevenlabs.io
Author: ElevenLabs editorial sitemap
Date: Published Oct 9, 2026

Context

Live transcription has two different outputs: words displayed while someone speaks, and completed segments retained as conversation history. ElevenLabs' Scribe v2 Realtime guide explains that early wording can change as more speech supplies . Applications can finalize segments themselves or let silence detection do it automatically. That choice determines when provisional text becomes a completed segment. In the guide's translation example, Scribe supplies the transcript and a separate browser translation API translates it.

Key takeaways · AI-distilled

ElevenLabs positions Scribe v2 for accuracy-focused batch transcription and Scribe v2 Realtime for ultra-low-latency uses like voice agents and meeting notetakers. Client-side should use a single-use generated server-side, not the API key.
Scribe v2 Realtime returns partial transcripts live and committed transcripts per segment. Commit can be manual, ideally at silences or turn boundaries, or automatic via voice activity detection when a silence threshold is reached.
The commit strategy matters because full-segment context improves accuracy: a partial "I scream" may read as "Ice cream" until the rest of the sentence arrives, so committed text gives a more accurate conversation history than partials.

Terms in this piece · Glossary

context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Read the source elevenlabs.io