Transcribe with word timestamps
openai/whisper-1Turn a recording into text with the start and end of every word.
- second
- $0.0001
- OpenAI
- 5 min
Overview
Takes an audio recording and returns what was said with a start and end time for every word, plus the segments the speech falls into. This is the endpoint for aligning a script to a recording: captions, word-level highlights, or cutting on a spoken word. The audio is sent as base64 and the gateway builds the upload; recordings are limited to 25 MB, so long material should be split first. A language code skips detection and improves accuracy; a short prompt can spell unusual names the way they should appear. Billing counts the seconds of audio the transcriber reports.
The transcript text, the language, the duration in seconds, every word with its start and end in seconds, and the segments.
Input
These go in the input object of the run request.
| Field | Type | Required | Description |
|---|---|---|---|
| audio_base64 | base64 | Yes | The recording, base64-encoded. |
| filename | string | No | A name for the uploaded file, which sets its extension. |
| content_type | string | No | Media type of the recording. |
| duration_seconds | number | No | Length of the recording, used for billing when the audio carries no readable header. |
| language | string | No | The spoken language as an ISO 639-1 code, such as en or ko. |
| prompt | string | No | Optional text that guides spelling and style, such as names that appear in the recording. |
Requests up to 35 MB, audio up to 25 MB.
Example
A call that works, with your own key.
curl https://gateway.agentsky.dev/v1/run \
-H "authorization: Bearer $AGENTSKY_API_KEY" \
-H "content-type: application/json" \
-d '{"provider":"openai","endpoint":"whisper-1","input":{"audio_base64":"UklGRiQAAABXQVZFZm10IBAAAAABAAEA...","filename":"take.wav","content_type":"audio/wav","language":"en"}}'Pricing
$0.0001 per second on the default plan. Every run carries its price in the response.