Skip to main content

openai/gpt-4o-transcribe-diarize/speech-to-text

OpenAI’s GPT-4o Transcribe Diarize transcribes a recording from an audio URL and tags each segment with a speaker label and start and end times, for files up to 25 MB and two hours long, with an optional language hint.

From $0.0001/s

Workbench

Input

Speech to text
1/1
1 / 1 files
Conversation to split into speaker turns, up to 25 MB at least 1 characters
Two-letter language of the recording at least 2 characters · at most 2 characters
Sampling temperature between 0 and 1 no lower than 0 · no higher than 1

Checking your account before enabling submission and pricing.

Output

Ready for a new run
Example result

Audio length: 0:02

The next step is ready, please continue.

Related models and next tasks

Missing the model you need?

CONVERSATIONS WITH SPEAKER TURNS

Follow a recorded exchange by voice, passage and time range.

Key numbers

Context / structure
A / B
Speaker labels separate utterances within a conversation so you can follow turns before assigning participant names yourself.
File limit
25 MB
Diarize accepts recordings no larger than 25 MB and no longer than two hours.
Audio limit
2 h
Diarize accepts recordings no larger than 25 MB and no longer than two hours.
Minimum billing
1 s
Measured audio seconds

Before you spend

All modes
  • Diarize labels distinguish voices in one recording; they do not establish legal identity or match people across files.
  • Crosstalk and rapid interruptions can confuse speaker boundaries, so attribution needs a listening check.
  • Diarize accepts recordings no larger than 25 MB and no longer than two hours.
  • This speaker-focused endpoint has no prompt, reference-voice or timestamp-granularity input; automatic chunking is applied for you.
  • OpenAI has deprecated GPT-4o Transcribe Diarize and will end its service on 2027-02-26; Diarize remains available under normal billing before shutdown. For new projects, use GPT Transcribe for ordinary transcription, and plan the migration of any speaker-labelling requirements before Diarize retires.

What it does

Alternating violet and lilac wave sections are joined by fine gold connectors.

Speaker turns

See who has the floor

Speaker labels separate utterances within a conversation so you can follow turns before assigning participant names yourself. Keep a label until a participant or reviewer confirms the attribution. Similar voices and interruptions deserve a listen before a named quotation is published.

Speaker turns

Return to a timed passage

Purple glass sound segments sit at different points along an unmarked brass rail.

Segments include start and end times alongside speech, giving a reviewer a place to resume listening. Use those intervals to seek to a reply or a change of turn in your own player. Listen around the boundary before choosing an edit point.

Three abstract purple resonant columns send separate gold arcs toward a central vessel.

Speaker turns

Longer exchanges, ordered segments

Automatic chunking handles longer recordings while the completed result keeps the conversation's segment order. The integration supplies automatic chunking rather than asking you to choose a segmentation setting. Preserve the returned segment order when assembling the conversation view.

Where it fits

  • Podcast selections

    Find an exchange and use its time range to audition a candidate edit.

  • Moderated interviews

    Separate the moderator's questions from the participant's responses.

  • Meeting review

    Read speaker turns while checking who introduced a proposal.

  • Research attribution

    Verify a participant's quotation before linking it to a finding.

Model reference

GPT-4o Transcribe Diarize API reference

GPT-4o Transcribe Diarize API specifications

Input
GPT-4o Transcribe Diarize · Audio URL · MP3 / MP4 / MPEG / MPGA / M4A / WAV / WEBM
Language
Optional ISO 639-1
Temperature
0–1; default 0
Billing
Measured audio seconds
Delivery
Asynchronous task

How to call the GPT-4o Transcribe Diarize API

  1. Test an alternating exchange

    Upload a brief conversation containing a clear change between two voices.

  2. Inspect the segments

    Read the labels and time ranges in order, listening at the first speaker transition.

  3. Confirm attribution

    Match labels to known participants only after checking the recording with a reviewer.

The short version

Read the labels and time ranges in order, listening at the first speaker transition. Match labels to known participants only after checking the recording with a reviewer.

GPT-4o Transcribe Diarize API questions

All GPT models18