Skip to main content

elevenlabs/scribe-v3/speech-to-text

ElevenLabs Scribe V3 transcribes recorded audio into text with word timings, language detection and optional speaker labels.

From $0.0002/s

Workbench

Input

Speech to text
1/1
1 / 1 files
The recording to transcribe. at least 1 characters

Checking your account before enabling submission and pricing.

Output

Ready for a new run
Example result
Model response

Related models and next tasks

Missing the model you need?

RECORDED SPEECH TO TEXT

Word-level transcripts with speaker labels across 102 languages.

Key numbers

Language codes
112
ElevenLabs Scribe V3 lists 112 language codes, and an empty selection is detected.
Longest file
1 h
ElevenLabs Scribe V3 accepts a recording of up to one hour.
Minimum bill
10s
ElevenLabs Scribe V3 bills per second of audio, with a 10-second minimum, and does not charge a failed request.
Extra hints
100
ElevenLabs Scribe V3 accepts up to 100 additional language codes you expect to hear.
Speaker labels
OFF
ElevenLabs Scribe V3 leaves speaker labels off until you enable them.
Event tags
ON
ElevenLabs Scribe V3 marks non-speech sounds unless you turn that tagging off.

What it does

OBSERVED TRANSCRIPT

A short note with word timings

Our November 18, 2026 run used a 2.35-second English recording. It returned “The next step is ready. Please continue.”, language code eng, and start and end times for every word and space. Use those timings to locate the wording in the source recording.

Waveform of a 2.35-second clip above the start and end time of each transcribed word.

LANGUAGE CONTEXT

Declare the languages you expect

Leave the spoken language empty for detection, or set a supported code when you know it. Add up to 100 other expected language codes as hints. Set this context for interviews, field recordings and talks where you know which languages to expect.

TRANSCRIPT PRESENTATION

Choose labels, events and numbers

The returned sentence with its language code, clip duration and a table of word timings.

Speaker labeling starts off, while audio event tagging starts on. Change those controls to suit an interview transcript or an editorial draft, then choose digits or words for spoken numbers. Review the resulting text and labels against the source recording.

Where it fits

  • Interview transcripts for editors

    Request speaker labels with diarize when organizing a recorded interview into speaking turns. Review the labels against the audio before assigning quotations to people.

  • Caption preparation for recorded talks

    Use word timings to locate speech while preparing captions in an editing workflow. Our English note returned start and end times for words and spaces; caption layout and final timing still need editorial review.

  • Language context for field recordings

    For recordings expected to contain more than one language, set the main spoken language and add the other expected codes as hints. Leave the main language empty when you want it detected.

  • Quantity checks in work interviews

    Choose digits when reviewing spoken quantities, or words when preparing prose. The number-format control sets the presentation; listen back to figures before carrying them into a report.

  • Short spoken progress notes

    Turn a brief recorded status message into text that can be reviewed alongside the audio. Our 2.35-second clip returned “The next step is ready. Please continue.” with detected language eng and timings for each word and space.

Model reference

ElevenLabs Scribe V3 API reference

ElevenLabs Scribe V3 API specifications

Publisher
ElevenLabs
Release date
November 10, 2026
Task
Recorded speech to text
Required input
audio_url: a nonempty recording URL
Language selection
ISO 639 code from the supported list; leave empty for detection
Additional language hints
Up to 100 expected language codes
Number formatting
digits or words
Speaker labels
diarize; off by default
Audio event tags
tag_audio_events; on by default
Recording length
Up to one hour
Billing
Per second of audio; 10-second minimum per request; failed requests are not charged
Observed output
2.35-second English clip; transcript, language code eng, and start/end timings for words and spaces

How to call the ElevenLabs Scribe V3 API

  1. Prepare the recording URL

    Provide audio_url for a recording of up to one hour. Listen to the source so you know which passages, speaker changes and numbers will need attention during review.

  2. Choose the language context

    Set language to a supported code when known, or leave it empty for detection. Add other expected languages through language_hints, with one code per row.

  3. Set transcript presentation

    Enable diarize if you need speaker labels, and decide whether to retain audio event tags. Select digits or words for numbers when your editing workflow needs a consistent form.

  4. Review text against the audio

    Submit the recording and compare the returned wording with the source, using word timings to find passages. Check speaker turns and important numbers before using the transcript in published work.

The short version

Bring a recorded interview, talk or spoken note and turn it into text you can review alongside the source. Choose the language context, enable speaker labels for conversations, and decide how numbers and sound events should appear. Our English example returned the note’s wording with a detected language and timings for words and spaces. Use those timings to locate passages, then check quotations and figures before publication. Recordings can run up to one hour; estimate the charge using the displayed per-second rate and the 10-second minimum per request.

ElevenLabs Scribe V3 API questions