AI-MODELBEATS

Custom Hugging Face Inference Endpoint handler for semantic audio labeling with BEATs.

Input

{
  "inputs": {
    "audio_url": "https://...",
    "video_url": "https://..."
  }
}

Provide either audio_url or video_url.

Output

{
  "audio_mood": "upbeat|emotional|dramatic|tense|calm",
  "energy_level": "low|medium|high",
  "voice_music_ratio": "balanced",
  "evidence": {
    "top_labels": [],
    "tempo_bpm": 0,
    "rms": 0
  },
  "model_id": "repo:file"
}

This endpoint is semantic-only. voice_music_ratio is intentionally left neutral so it can be paired with a dedicated speech/music endpoint.