Skip to main content

openai/gpt-live-transcribe/realtime-transcription

GPT Live Transcribe emits text while audio arrives, with language hints, vocabulary context and adjustable transcript delay.

From $0.00006/s80% below the publisher's list price

Workbench

Input

Realtime transcription
0/2048
Give Live Transcribe a concise description of the recording’s subject, without inventing the words being spoken. at most 2048 characters
1/1
1 / 1 files
For one Live Transcribe task, provide a short WAV containing 24 kHz mono PCM or 8 kHz G.711; live audio uses the transcription WebSocket. at least 1 characters
affects price
Live Transcribe admits each segment for this maximum duration, from 1 to 60 seconds; commit earlier when a complete phrase is ready. no lower than 1 · no higher than 60
Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

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?

GPT Live Transcribe emits text while audio arrives, with language hints, vocabulary context and adjustable transcript delay.

Key numbers

Segment maximum
60s

What it does

Captions before the speaker finishes

A tunable delay controls how long GPT Live Transcribe waits before emitting transcript text. A shorter delay can make a caption display feel more immediate, while a larger delay gives the model more speech to consider before showing a new fragment. Treat intermediate text as provisional in the interface, then use the committed segment’s final transcript for downstream processing. Free-form context can describe a meeting topic or a product discussion, while keyword hints can identify names that ordinary vocabulary does not cover well. Multiple language hints help with multilingual speech and code-switching. These are recognition aids, not commands to replace what the speaker said. Live Transcribe produces text rather than a spoken answer and does not provide speaker labels, word timestamps or confidence scores.

Where it fits

  • Live webinar captions

    For a live webinar caption panel using GPT Live Transcribe, send short microphone chunks and keep the current segment visually distinct from finalized captions. Commit at an application-controlled boundary so an editor can associate final text with the correct portion of the session.

  • Terminology-aware operator console

    For an operator console, supply the expected product names and likely languages before speech begins. Let the operator review completed segments before they become a stored case note or trigger an application workflow.

Model reference

GPT Live Transcribe API reference

How to call the GPT Live Transcribe API

  1. Prompt guide & Pro tips

    Describe the subject of the audio to GPT Live Transcribe in a short context note, then list relevant terms with their normal spelling. Avoid supplying a guessed full transcript: it can bias recognition toward words that were never spoken. If two languages are likely to alternate, provide both hints and check names, numbers and sentence boundaries with representative audio. Choose a delay profile using real listening conditions, including brief pauses and background noise. This connection uses explicit commits, not automatic turn detection; select a segment limit between one and sixty seconds and commit before that limit when a natural boundary occurs. PCM must be 24 kHz, 16-bit mono, or use 8 kHz G.711. A short matching WAV can exercise one segment in the task interface. Use the final reported audio duration for billing, and retain only the transcript your application actually needs.

GPT Live Transcribe API questions

All GPT models18