parameters.json

{
    "features" : {
        "sequence_time": 1.0, 
        "sequence_hop_time":0.5,
        "audio_hop":680,
        "audio_win":1024,
        "sr": 22050,
        "Spectrogram" : {
            "n_fft":1024
        },
        "MelSpectrogram" : {
            "mel_bands": 64,
            "n_fft":1024
        },
        "Openl3" : {
            "content_type" : "env",
            "input_repr" : "mel256", 
            "embedding_size" : 512
        },
        "RawAudio" : {},
        "FramesAudio" : {}
    },

    "train" : {
        "early_stopping" : 30,
        "epochs" : 50,
        "considered_improvement" : 0,
        "learning_rate" : 0.001,
        "batch_size" : 128,
        "verbose" : 1,
        "optimizer" : "Adam"
    },

    "models" : {
        "MLP" : {
            "normalizer" : "minmax",
            "model_arguments" : {},
            "train_arguments": {}
        },
        "SB_CNN" : {
            "normalizer" : "minmax",  
            "model_arguments" : {
                "n_dense_cnn" : 64,
                "n_channels" : 0,
                "filter_size_cnn" : [5, 5], 
                "pool_size_cnn" : [2,2]
            },
            "train_arguments": {}
        },
        "SB_CNN_SED" : {
            "normalizer" : "minmax",     
            "model_arguments" : {},
            "train_arguments": {}
        },
        "A_CRNN" : {
            "normalizer" : "minmax",     
            "model_arguments" : {},
            "train_arguments": {}
        },     
        "VGGish" : {
            "normalizer" : "minmax",     
            "model_arguments" : {},
            "train_arguments": {}
        } 
    },

    "datasets" : {
        "UrbanSound8k" : {
            "dataset_path" : "datasets/UrbanSound8K",
            "evaluation_mode" : "cross-validation"
        },
        "ESC50" : {
            "dataset_path" : "datasets/ESC50",
            "evaluation_mode" : "cross-validation"
        },
        "ESC10" : {
            "dataset_path" : "datasets/ESC50",
            "evaluation_mode" : "cross-validation"
        },
        "URBAN_SED" : {
            "dataset_path" : "datasets/URBAN-SED_v2.0.0",
            "evaluation_mode" : "train-validate-test"
        },
        "SONYC_UST" : {
            "dataset_path" : "datasets/SONYC_UST",
            "evaluation_mode" : "train-validate-test"
        },
        "TAUUrbanAcousticScenes2020Mobile" : {
            "dataset_path" : "datasets/TAUUrbanAcousticScenes2020Mobile",
            "evaluation_mode" : "train-test"        
        },
        "TAUUrbanAcousticScenes2019" : {
            "dataset_path" : "datasets/TAUUrbanAcousticScenes2019",
            "evaluation_mode" : "train-test"          
        },
        "TUTSoundEvents2017" : {
            "dataset_path" : "datasets/TUTSoundEvents2017",
            "evaluation_mode" : "cross-validation-with-test"          
        },
        "FSDKaggle2018" : {
            "dataset_path" : "datasets/FSDKaggle2018",
            "evaluation_mode" : "train-validate-test"          
        },
        "MAVD" : {
            "dataset_path" : "datasets/MAVD",
            "evaluation_mode" : "train-validate-test"          
        }  
    },
    "data_augmentations": [
        {"type": "pitch_shift", "n_semitones": -1},
        {"type": "time_stretching", "factor": 1.05},
        {"type": "white_noise", "snr": 60}  
    ]
}