Name the sound in one 5-second window of a generated test clip

Arguments

{
  "source": "$CLIP",
  "start_sec": 0,
  "end_sec": 5,
  "window_sec": 5,
  "labels": [
    "siren",
    "church bells",
    "car horn",
    "dog barking",
    "clapping",
    "train",
    "helicopter",
    "glass breaking",
    "thunderstorm",
    "rooster crowing"
  ]
}

Result

{
  "ok": true,
  "data": {
    "source_path": "~/Code/agentic-video-gallery/cards/label-synthetic/clip.mp4",
    "start_sec": 0,
    "end_sec": 5,
    "window_sec": 5,
    "model": "clef-omni",
    "labels": [
      "siren",
      "church bells",
      "car horn",
      "dog barking",
      "clapping",
      "train",
      "helicopter",
      "glass breaking",
      "thunderstorm",
      "rooster crowing"
    ],
    "windows": [
      {
        "start_sec": 0,
        "end_sec": 5,
        "top_label": "siren",
        "top_chance": 0.7629,
        "none_chance": 0.035,
        "chances": {
          "siren": 0.7629,
          "church bells": 0.0197,
          "car horn": 0.0407,
          "dog barking": 0.0168,
          "clapping": 0.0136,
          "train": 0.0223,
          "helicopter": 0.031,
          "glass breaking": 0.0155,
          "thunderstorm": 0.0234,
          "rooster crowing": 0.0191
        }
      }
    ],
    "input_tokens": 385,
    "complete": true,
    "next_start_sec": null,
    "note": "Audio was sent to Cloudflare Workers AI. Chances are the model's own and sum to 1 with none_chance; top_label is null when 'none of these' is more likely than every label. Sounds outside `labels` are forced onto the nearest label or none, so give every sound that may be present."
  }
}