Serverless Transcriber

RunPod serverless endpoint for GPU-accelerated audio transcription with speaker diarization.

Models used:

• WhisperX (whisper base.en) — transcription
• PyAnnote (speaker-diarization-3.1) — speaker labels

API

POST /run

{
  "input": {
    "audio_url": "https://example.com/audio.mp3",
    "model": "base.en",
    "diarize": true
  }
}

Parameters:

• audio_url — direct URL to audio file (mp3, wav, m4a, etc.)
• audio_data — base64-encoded audio (alternative to URL)
• model — whisper model: tiny.en, base.en, small.en, medium, large-v3
• diarize — boolean, enable speaker identification

Response:

{
  "transcript": "[00:00] [SPEAKER_00] Hello world",
  "segments": [
    {"start": 0.0, "end": 2.5, "speaker": "SPEAKER_00", "text": "Hello world"}
  ],
  "language": "en",
  "model": "base.en",
  "duration_seconds": 42.0
}