{
    "name": "EffnetDiscogs",
    "type": "Music embeddings based on contrastive similarity",
    "link": "https://essentia.upf.edu/models/feature-extractors/discogs-effnet/discogs_track_embeddings-effnet-bs64-1.pb",
    "version": "1",
    "description": "Prediction of track-level similarity using the bilinear similarity",
    "author": "Pablo Alonso",
    "email": "pablo.alonso@upf.edu",
    "release_date": "2022-06-15",
    "framework": "tensorflow",
    "framework_version": "2.8.0",
    "classes": [],
    "model_types": [
        "frozen_model",
        "onnx"
    ],
    "dataset": {
        "name": "Discogs-4M (unreleased)",
        "citation": "In-house dataset",
        "size": "3.3M full tracks (3.3M used)",
        "metrics": {
            "accuracy (validation test)": 0.89
        }
    },
    "schema": {
        "inputs": [
            {
                "name": "serving_default_melspectrogram",
                "type": "float",
                "shape": [
                    64,
                    128,
                    96
                ]
            }
        ],
        "outputs": [
            {
                "name": "PartitionedCall:0",
                "type": "float",
                "shape": [
                    64,
                    512
                ],
                "op": "fully connected",
                "description": "track similarity",
                "output_purpose": "predictions"
            },
            {
                "name": "PartitionedCall:1",
                "type": "float",
                "shape": [
                    64,
                    1280
                ],
                "op": "Flatten",
                "output_purpose": "embeddings"
            }
        ]
    },
    "citation": "@inproceedings{alonso2022music,\n  title={Music Representation Learning Based on Editorial Metadata from Discogs},\n  author={Alonso-Jim{\\'e}nez, Pablo and Serra, Xavier and Bogdanov, Dmitry},\n  booktitle={Conference of the International Society for Music Information Retrieval (ISMIR)},\n  year={2022}\n}",
    "inference": {
        "sample_rate": 16000,
        "algorithm": "TensorflowPredictEffnetDiscogs"
    }
}