# Whisper Diarization - Core.Today AI API

> Whisper transcription with speaker diarization (8M+ runs) — returns who said what, with per-segment speaker labels and timestamps. The go-to for meetings and interviews.

- **Provider**: Whisper
- **Model ID**: thomasmol/whisper-diarization
- **Category**: Audio & TTS
- **Credits**: 7 per run
- **Speed**: Fast
- **Quality**: High

## Features
- Speaker-labeled transcription (who said what)
- Per-segment timestamps
- Auto speaker-count detection (or specify)
- Vocabulary prompt for names and jargon
- English translation option

## Use Cases
- Meeting minutes with speaker attribution
- Interview transcription
- Podcast multi-host transcripts
- Call center conversation analysis
- Court/debate record digitization

## API Endpoint

Base URL: https://api.core.today/v1
Create Prediction: POST /predictions
Get Status: GET /predictions/{job_id}

## Authentication
Header: X-API-Key: YOUR_API_KEY

## Input Parameters

### Optional
- **prompt**: string - Names, acronyms, and loanwords to improve accuracy
- **file_string**: string - Either provide: Base64 encoded audio file,
- **file_url**: string - Direct audio file URL (use this, or 'file' upload)
- **file**: string - Or an audio file
- **num_speakers**: integer - Number of speakers. Leave empty to auto-detect
- **translate**: boolean (default: false) - Translate the speech into English
- **language**: string - Language code like 'ko', 'en'. Empty auto-detects

## Examples

### Two-Person Interview
Transcribe with speaker labels for two speakers
```json
{
  "model": "thomasmol/whisper-diarization",
  "input": {
    "file_url": "https://example.com/interview.mp3",
    "num_speakers": 2,
    "language": "ko",
    "prompt": "코어닷투데이, API, 크레딧"
  }
}
```

## Response Format
```json
{
  "job_id": "abc123",
  "status": "pending | processing | completed | failed",
  "result": "URL or data (when completed)"
}
```

## Usage Flow
1. POST /predictions with model and input -> receive job_id
2. GET /predictions/{job_id} -> poll until status is completed or failed
3. Result contains output URL(s)

## Tips
- Specify num_speakers when known — auto-detection can over-split similar voices
- Put participant names and product terms in prompt for cleaner transcripts
- Result segments carry speaker, start/end times, and text — ready for meeting-minute UIs
- Cheapest of the three Whisper options (9 credits)

## Documentation
https://replicate.com/thomasmol/whisper-diarization