config/facodec.json

{
    "exp_name": "facodec",
    "model_type": "FAcodec",
    "log_dir": "./runs/",
    "log_interval": 10,
    "save_interval": 1000,
    "device": "cuda",
    "epochs": 1000,
    "batch_size": 4,
    "batch_length": 100,
    "max_len": 80,
    "pretrained_model": "",
    "load_only_params": false,
    "F0_path": "modules/JDC/bst.t7",
    "dataset": "dummy",
    "preprocess_params": {
        "sr": 24000,
        "frame_rate": 80,
        "duration_range": [1.0, 25.0],
        "spect_params": {
            "n_fft": 2048,
            "win_length": 1200,
            "hop_length": 300,
            "n_mels": 80,
        },
    },
    "train": {
        "gradient_accumulation_step": 1,
        "batch_size": 1,
        "save_checkpoint_stride": [20],
        "random_seed": 1234,
        "max_epoch": -1,
        "max_frame_len": 80,
        "tracker": ["tensorboard"],
        "run_eval": [false],
        "sampler": {"holistic_shuffle": true, "drop_last": true},
        "dataloader": {"num_worker": 0, "pin_memory": true},
    },
    "model_params": {
        "causal": true,
        "lstm": 2,
        "norm_f0": true,
        "use_gr_content_f0": false,
        "use_gr_prosody_phone": false,
        "use_gr_timbre_prosody": false,
        "separate_prosody_encoder": true,
        "n_c_codebooks": 2,
        "timbre_norm": true,
        "use_gr_content_global_f0": true,
        "DAC": {
            "encoder_dim": 64,
            "encoder_rates": [2, 5, 5, 6],
            "decoder_dim": 1536,
            "decoder_rates": [6, 5, 5, 2],
            "sr": 24000,
        },
    },
    "loss_params": {
        "base_lr": 0.0001,
        "warmup_steps": 200,
        "discriminator_iter_start": 2000,
        "lambda_spk": 1.0,
        "lambda_mel": 45,
        "lambda_f0": 1.0,
        "lambda_uv": 1.0,
    },
}