ElevenLabs Scribe v2

What it does

ElevenLabs Scribe v2 takes audio in and produces text out. You give it a video or audio file and it returns a transcript with timestamps and speaker labels, so each line of text is attached to a point in the recording and to whoever was talking. The Transcription pipeline it powers covers 99+ languages. It does not generate audio, images, or video. The only output is the transcript and its associated timing and speaker metadata.

When to use it

• Running the Transcription pipeline on a recorded interview or podcast where you need to know who said which line, not just what was said.
• Turning a video file into text you can search, quote, or edit. The timestamps let you map any sentence back to the moment in the source.
• Transcribing non-English or mixed-language recordings, since the pipeline covers 99+ languages.
• Producing a timed, speaker-labelled transcript you can hand off to whatever comes next in your own workflow.
• Any job where the deliverable is text derived from speech, rather than something you need re-voiced or re-rendered.

What to know before you run it

Cost scales with the length of the audio. The wholesale units are per-minute, with 30-minute and 60-minute tiers, so a long recording costs proportionally more than a short one. Trimming to the section you actually need is the main lever you have. There is no quality or resolution setting to trade off, and no reference image or extra input: you upload one media file and get one transcript back. Speaker labels come from the model itself rather than from a setting you configure, so there is nothing to tune there. Check the labels against your recording if speaker attribution matters to the deliverable. There are no examples published for this model yet.

Also known as

You may see this model referred to as ElevenLabs Scribe, Scribe v2 (Batch), or by the identifier scribe_v2. These all refer to the same model on Pipe2.