Whisper-1 — Turn recordings into text
Turn recordings into searchable text or captions. Try Whisper-1 with your own audio and choose the output for your next step.
- Transcription
- English translation
- Word timestamps
- Subtitle files
Try Whisper-1 with a recording
Start with a short clip. Compare the transcript with the recording before publishing captions.
Choose your Whisper-1 output
The same recording can serve different workflows. These are format illustrations, not a transcription accuracy benchmark.
Welcome to the recording.Read and edit a plain transcript.
{ "word": "Welcome",
"start": 0.0, "end": 0.5 }Locate a word or a passage in the recording.
1
00:00:00,000 --> 00:00:02,000
Welcome to the recording.Import timed captions into your video editor.
Whisper-1 pricing
Pay for successfully transcribed audio at the current Whisper-1 rate.
Audio seconds are rounded up before conversion to billing units. A partial second can change the final charge. Transcript length does not determine the price.
| Model | Metering | Rate | Unit |
|---|---|---|---|
| Whisper-1 | One minute corresponds to 1,000 duration-based units. These are not words or text tokens. | $36.00 | Per million audio metering units |
Failed Whisper-1 requests are not charged.
What is Whisper-1?
Whisper-1 is OpenAI’s speech recognition model for turning recorded audio into text. Make interviews searchable, prepare captions, or bring spoken material into a written workflow.
Whisper-1 preserves the spoken language or translates speech into English. Choose readable text or timed output.
Recorded audio · Multilingual transcription · English translation
Use this exact ID in the model field.
Two ways to use Whisper-1
Build it into your application
Upload a file, save its task ID and retrieve the transcript. Specify Whisper-1 explicitly in your application.
- 1Choose the recording and parameters.
- 2Submit once and save the task ID.
- 3Download and review the transcript.
Work with your coding agent
Give your agent the audio path and Whisper-1 documentation. Review the request and estimated cost, then save the selected output.
- 1Copy this integration prompt.
- 2Provide your audio file and approve the request.
- 3Review the saved result.
From a Whisper-1 transcript to useful work
Readable notes
Turn interviews and voice memos into drafts you can search. Check names and specialist terms in the output before quoting.
Captions you can move
Import captions into your editor as SRT or VTT. Adjust line breaks and timing to suit the screen.
A place for every word
Locate phrases with word timestamps. Check alignment against the audio where precise cuts matter.
Put Whisper-1 to work
Interview review
Search a transcript for a topic, then return to the recording to confirm the quotation.
Lesson preparation
Transcribe your own lesson. Correct terminology and add paragraph breaks before sharing it with learners.
English working copies
Translate a recording so collaborators can read it in English. Retain the audio for review.
Start with Whisper-1 in four steps
- 01
Choose a recording
Use a supported file below the upload limit. A short, clear excerpt makes your first Whisper-1 result easier to check.
- 02
Choose the output
Keep the spoken language or translate to English. Choose text, timed JSON or captions.
- 03
Submit once
Save the task ID and check it until completion. A slow status check does not require another submission.
- 04
Review and download
Review the Whisper-1 result against the recording. Correct names, numbers and quotations before publishing.
Connect transcription to your application
Upload a file with multipart form data, then check the returned task ID. The Whisper-1 documentation lists every field, output format and request example.
curl https://api.seedrouter.ai/v1/audio/transcriptions \
-H "Authorization: Bearer $SEEDROUTER_API_KEY" \
-F model=whisper-1 \
-F [email protected] \
-F response_format=verbose_json \
-F 'timestamp_granularities[]=word'curl "https://api.seedrouter.ai/v1/tasks/$TASK_ID" \
-H "Authorization: Bearer $SEEDROUTER_API_KEY"| Parameter | Type | Description |
|---|---|---|
| file | Binary | One uploaded recording. |
| model | String | whisper-1 |
| language | String | Optional source language; transcription only. |
| prompt | String | Optional spelling and context. |
| response_format | Enum | json, text, srt, verbose_json, vtt. |
| temperature | Number | 0–1; default 0. |
| timestamp_granularities[] | Array | word / segment; verbose_json only. |
| chunking_strategy | String / object | Optional; compatibility remains unverified. |
| stream | Boolean | Ignored by Whisper-1. |
If a request fails
Check the file, parameters and balance. If submission is uncertain, check task history before submitting again.
Which Whisper-1 format fits?
Choose the Whisper-1 format that fits your next step.
| Your next step | Choose | What you get |
|---|---|---|
| Read or edit | text | Plain transcript |
| Use in an application | json | Transcript text in JSON |
| Find exact passages | verbose_json | Transcript with timestamp detail |
| Edit video captions | srt | Numbered subtitle cues |
| Caption a web player | vtt | WebVTT subtitle cues |
Know the limits of Whisper-1
Review before publishing
Whisper-1 can mishear speech or invent text, especially with difficult audio. Check names, quotations and consequential details against the original.
No speaker labels
Whisper-1 does not assign speaker identities. Word timestamps tell you when speech occurs; they do not tell you who spoke.
Recorded files, not live speech
Whisper-1 processes uploaded recordings. It does not provide live partial transcripts, and English translation does not generate a dubbed audio track.
Use Whisper-1 on SeedRouter
Check the price first
See the current rate and an estimate when your browser can read the audio duration.
Keep the output you need
Download the selected transcript or captions. Keep a copy alongside your recording.
Failed requests are not charged
A failed task does not use your balance. Check task history for its final status before starting another attempt.
Explore the model catalog
Turn a reviewed transcript into a summary with another model. Browse capabilities and live prices.
Whisper-1 questions
Can Whisper-1 translate audio?+
Whisper-1 translation produces English text; transcription preserves the spoken language. Neither creates audio. Check the original when precise wording matters.
Does Whisper-1 support speaker diarization?+
Whisper-1 timestamps locate speech, not speakers. Add speaker names during review or choose a model that supports diarization.
How should I prepare a long Whisper-1 recording?+
Compress larger recordings or split at natural pauses to meet the upload limit. Retain their order and review the joins.
Can I provide spelling hints to Whisper-1?+
Provide relevant names and context. Whisper-1 considers its final 224 prompt tokens; check the spelling in the returned transcript.
Which Whisper-1 output has word timestamps?+
Whisper-1 word timestamps require verbose_json. SRT and VTT provide subtitle cues; plain text and JSON suit transcript editing.
Is the Whisper-1 API identical to the synchronous OpenAI response?+
SeedRouter returns a task ID, followed by a transcript download URL. Synchronous SDKs need adaptation; see the reference for compatibility limits.
Hear it once. Work with the words.
Choose a recording, check the rate and try Whisper-1. Download your preferred format and review it beside the audio.