model.json

{
  "name": "MERIT",
  "task": "Disentangled music audio similarity",
  "source_repository": "AMAAI-Lab/MERIT",
  "source_revision": "aa0039b4ead7d451b5af9a9f3b4b7c206feaed51",
  "source_license": "MIT",
  "backbone": {
    "repo": "m-a-p/MERT-v1-330M",
    "revision": "5240c2708a5acaee1007f43fb9735c7dcd0b78c9",
    "license": "CC BY-NC 4.0"
  },
  "projection_heads": {
    "repo": "amaai-lab/merit",
    "revision": "a85df30eca1ba112eb594285f3ba1d96488e7883",
    "files": [
      "head_mel/best_head.pt",
      "head_rhy/best_head.pt",
      "head_tim/best_head.pt"
    ],
    "license": "MIT"
  },
  "input": {
    "sample_rate": 24000,
    "channels": 1,
    "minimum_duration_seconds": 5,
    "maximum_duration_seconds": 10,
    "embedding_window_seconds": 10
  },
  "output": {
    "type": "cosine_similarity",
    "range": [
      -1.0,
      1.0
    ],
    "dimensions": [
      "melody",
      "rhythm",
      "timbre"
    ]
  }
}