# Muse Voice Transcribe 1.0 - Core.Today AI API

> Meta's Muse Voice Transcribe speech-to-text over the Meta developer API - 25 languages with code-switching (Korean included), speaker diarization, endpointing, keyword and language biasing. Billed per second of audio (0.11625 credits/s, about 7 credits per minute), up to 10 minutes per request.

- **Provider**: Meta
- **Model ID**: meta/muse-voice-transcribe-1.0
- **Category**: Audio & TTS
- **Credits**: 7 per minute of audio (0.11625/s, settled on measured length)
- **Speed**: Fast
- **Quality**: High

## Features
- 25 languages with code-switching, Korean included
- Speaker diarization and endpointing (turns with speaker labels and timestamps)
- Keyword biasing for names, acronyms and product terms
- Language hints steer recognition without forcing it
- Accepts mp3/m4a/wav/ogg/webm - normalized to 24 kHz mono WAV before upload

## Use Cases
- Meeting and interview transcription with speaker labels
- Call recordings and voice memos in mixed Korean/English
- Podcast and lecture transcripts for search and notes

## 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

### Required
- **audio**: string - Audio file URL (mp3, m4a, wav, ogg, webm, mp4 audio track). Normalized to mono 16-bit PCM WAV at 24 kHz before upload. Max 10 minutes per request.

### Optional
- **mode**: string (default: PUSH_TO_TALK) - PUSH_TO_TALK returns one transcript; ENDPOINTING splits into turns at speech boundaries; DIARIZATION additionally labels speakers.
  Options: PUSH_TO_TALK, ENDPOINTING, DIARIZATION
- **keywords**: array - Names, acronyms or product terms to bias recognition toward.
- **language_bias**: array - Language hint (steers recognition, does not force it). Omit for automatic detection across all 25 languages.
  Options: Arabic, Bengali, Dutch, English, French, German, Hebrew, Hindi, Indonesian, Italian, Japanese, Kannada, Korean, Malay, Mandarin Chinese, Marathi, Polish, Portuguese, Spanish, Tagalog, Tamil, Telugu, Thai, Turkish, Vietnamese

## Examples

### Transcribe a recording
Returns a single transcript for an English recording.
```json
{
  "model": "meta/muse-voice-transcribe-1.0",
  "input": {
    "audio": "https://replicate.delivery/mgxm/e5159b1b-508a-4be4-b892-e1eb47850bdc/OSR_uk_000_0050_8k.wav",
    "mode": "PUSH_TO_TALK",
    "language_bias": [
      "English"
    ]
  }
}
```

### Meeting with speaker labels
DIARIZATION mode returns turns with speaker, startMs and endMs; keywords bias product names.
```json
{
  "model": "meta/muse-voice-transcribe-1.0",
  "input": {
    "audio": "https://example.com/meeting.m4a",
    "mode": "DIARIZATION",
    "keywords": [
      "Core.Today",
      "Seedance"
    ],
    "language_bias": [
      "Korean",
      "English"
    ]
  }
}
```

## 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
- The result is JSON: transcript, audioDurationMs (the billed length) and, in ENDPOINTING/DIARIZATION modes, a turns array
- Word-level timestamps and confidence scores are not available; use turns for segment timing
- Split recordings longer than 10 minutes before submitting - longer files are rejected
- Live/streaming transcription (the realtime WebSocket endpoint) is not exposed through this API

## Documentation
https://dev.meta.ai/docs/speech-to-text