Whisper turns spoken audio into accurate written text. Upload an audio file URL and receive a clean transcription, with optional timestamped subtitle output (SRT or VTT) for video captioning, podcast transcripts, meeting notes, and voice-driven workflows.
About this model
Whisper is a robust speech-to-text model that converts spoken audio into accurate written transcripts across dozens of languages. Submit an audio URL and receive a clean transcription, optional segment-level timestamps, or ready-to-use SRT/VTT subtitles, making it ideal for podcasts, meetings, video captioning, and any voice-driven workflow.
Cost analysis
| Provider | Cost | Notes |
|---|---|---|
| muapiapp | $0.012 per minute | Pay only for the audio you transcribe, billed per minute with a 1-minute minimum. |
| Fal.ai | Not available | Whisper is not offered as a managed endpoint on Fal.ai. |
| Replicate | ~$0.011 per minute | Pricing varies with the underlying GPU; muapiapp offers a flat, predictable per-minute rate. |
Pay only for the audio you transcribe, billed per minute with a 1-minute minimum.
Whisper is not offered as a managed endpoint on Fal.ai.
Pricing varies with the underlying GPU; muapiapp offers a flat, predictable per-minute rate.
* Competitor pricing is estimated based on similar model architectures and usage tiers.
Configuration schema
| Parameter | Type | Description | Default |
|---|---|---|---|
| Audio URL | string | URL of the audio file to transcribe. Supported formats: mp3, mp4, mpeg, mpga, m4a, wav, webm. File must be under 25 MB. | https://d3adwkbyhxyrtq.cloudfront.net/ai-audio/sample.mp3 |
| Language | string | Optional ISO-639-1 language code of the input audio (e.g. 'en', 'es', 'hi'). Leave empty for automatic detection. | en |
| Prompt | string | Optional context to guide the model's style or to spell out unusual words and proper nouns. Should match the audio language. | The transcript is a podcast about AI startups. |
| Response Format | Enum (5 options) | Output format. 'json' / 'text' return plain transcripts, 'srt' / 'vtt' return timestamped subtitles, 'verbose_json' includes per-segment metadata. | json |
| Temperature | number | Sampling temperature between 0 and 1. Higher values make output more random; lower values make it more deterministic. | 0 |
URL of the audio file to transcribe. Supported formats: mp3, mp4, mpeg, mpga, m4a, wav, webm. File must be under 25 MB.
https://d3adwkbyhxyrtq.cloudfront.net/ai-audio/sample.mp3Optional ISO-639-1 language code of the input audio (e.g. 'en', 'es', 'hi'). Leave empty for automatic detection.
enOptional context to guide the model's style or to spell out unusual words and proper nouns. Should match the audio language.
The transcript is a podcast about AI startups.Output format. 'json' / 'text' return plain transcripts, 'srt' / 'vtt' return timestamped subtitles, 'verbose_json' includes per-segment metadata.
jsonSampling temperature between 0 and 1. Higher values make output more random; lower values make it more deterministic.
0Developer documentation
Provide an Audio URL
audio_url. Supported formats include mp3, mp4, mpeg, mpga, m4a, wav, and webm. Files must be under 25 MB.Pick a Response Format
json / text return a plain transcript.srt and vtt return timestamped subtitle files ready to drop into a video editor.verbose_json exposes per-segment metadata (start, end, text, language).(Optional) Tune Recognition
language to an ISO-639-1 code (e.g. en, es, hi) to skip auto-detection and improve accuracy.prompt to bias the transcription toward specific spelling, jargon, or style.temperature (0.0 - 1.0) — lower is more deterministic, higher is more varied.Submit and Retrieve
Frequently asked
Whisper transcribes audio into written text. It supports many languages and can also produce timestamped subtitle formats.
mp3, mp4, mpeg, mpga, m4a, wav, and webm. Each file must be under 25 MB.
Billing is based on the audio duration, rounded up to the next minute. A 1-minute minimum applies to every request.
Yes — set `response_format` to `srt` or `vtt` to receive a ready-to-use subtitle file with start and end timestamps for every segment.
Yes. Leave the `language` field empty for auto-detection, or specify an ISO-639-1 code to lock the recognition to a known language for better accuracy.