# Scribe v3 for recorded and live speech

_Release #405, November 11, 2026. Voice, Agent smarts_

Viktor now uses ElevenLabs Scribe v3 for recorded speech and Scribe v3 Live for realtime transcription.

![Hand-drawn sketch of Scribe v3 for recorded audio and Scribe v3 Live for live speech, with number-format and language options.](https://viktor.com/changelog/2026-11-11-scribe-v3/hero@2x.webp)

- **A new transcription default.** Viktor now uses ElevenLabs Scribe v3 for recorded speech and Scribe v3 Live for realtime transcription.
- **Numbers your way.** Ask for numbers written as digits or words, and specify secondary languages when transcribing multilingual audio.

## What's new

Recorded speech and live transcription now use the new Scribe family from ElevenLabs. Scribe v3 replaces the previous default for audio files, voice messages and recorded meetings. Scribe v3 Live takes over realtime meeting transcripts, speech in Spaces and the voice shortcut's live stream.

There are also two new options for recorded audio: how numbers are written, and which secondary languages to expect alongside the main language. That gives you more control when a recording switches languages or contains figures you want to use in a report.

## How to use it

Send an audio recording and ask Viktor to transcribe it, or keep using your existing speech workflow. The new models are the defaults, so there is nothing to switch on. For recorded audio, tell him whether you want digits or words for numbers, and name the main and secondary languages.

## Where it works

Audio files sent in web and desktop chat, recorded meetings, live meeting transcription, voice messages, Spaces speech and the voice shortcut use the updated defaults. Existing workflows that explicitly request the previous Scribe model can still use it.

Source: https://viktor.com/changelog/2026-11-11-scribe-v3
