Workbench
Output
Ready for a new runRelated models and next tasks
Missing the model you need?
CLASSIC SPEECH RECOGNITION, TIMED TEXT
Bring recorded words onto an editable caption timeline.
Key numbers
- Context / structure
- word
- Word intervals let a transcript viewer seek near a selected term and highlight speech during playback.
- File limit
- 25 MB
- Whisper's word boundaries are approximate; pauses, music and simultaneous voices can shift an interval.
- Audio limit
- 2 h
- Whisper's word boundaries are approximate; pauses, music and simultaneous voices can shift an interval.
- Minimum billing
- 1 s
- Measured audio seconds
Before you spend
All modes
- Whisper's word boundaries are approximate; pauses, music and simultaneous voices can shift an interval.
- The caption workflow accepts at most 25 MB and 7200 seconds of recorded media per task.
- Word and segment timing are configured for this endpoint; selectable subtitle formats and translation are not public inputs here.
- A vocabulary prompt is recognition context, not a general instruction channel for designing subtitle style.
- OpenAI has announced that whisper-1 is deprecated and its API service will stop on 2027-02-26; until retirement, Whisper 1 continues to work and incur its usual charges. Begin new file-transcription integrations with GPT Transcribe instead of the retiring Whisper API.
What it does

Word timing
Navigate by individual words
Word intervals let a transcript viewer seek near a selected term and highlight speech during playback. Use these intervals to build a player that jumps from a selected word to the recording. Alignment is approximate, especially near pauses or overlapping voices.
Word timing
Arrange longer caption passages

Segment intervals provide a first structure for subtitles; you choose readable line breaks after transcription. Treat the intervals as an editing starting point, then choose line breaks and reading time yourself. Timing arrays do not by themselves constitute a finished subtitle file.

Word timing
Context for an archive excerpt
A short prompt can supply uncommon spellings or preceding speech to the API version of the classic open-source Whisper recognizer. For a recording split into parts, preceding text can provide continuity in the next request. Subtitle styling and final punctuation remain part of your editing workflow.
Where it fits
Caption preparation
Build a draft timeline, then adjust subtitles against the video.
Language practice
Replay a phrase and inspect the word intervals around a difficult term.
Archive navigation
Find a spoken phrase in an older recording without scrubbing the entire file.
Transcript players
Connect a selected word or segment to a playback position.
Model reference
Whisper 1 API reference
Whisper 1 API specifications
- Input
- Whisper 1 · 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 Whisper 1 API
Pick speech with a pause
Try an excerpt with an audible boundary to inspect Whisper's alignment.
Supply concise context
Add relevant spellings or the preceding passage in the same language as the audio.
Edit after alignment
Keep the original timing arrays while formatting and correcting the caption draft.
The short version
Add relevant spellings or the preceding passage in the same language as the audio. Keep the original timing arrays while formatting and correcting the caption draft.
