{
    "base_config": "config/ns2.json",
    "model_type": "NaturalSpeech2",
    "dataset": [
        "libritts"
    ],
    "preprocess": {
        "use_mel": false,
        "use_code": true,
        "use_spkid": true,
        "use_pitch": true,
        "use_duration": true,
        "use_phone": true,
        "use_len": true,
        "use_cross_reference": true,
        "train_file": "train.json",
        "valid_file": "test.json",
        "melspec_dir": "mel",
        "code_dir": "code",
        "pitch_dir": "pitch",
        "duration_dir": "duration",
        "metadata_dir": "metadata",
        "read_metadata": true,
        "clip_mode": "start"
    },
    "model": {
        "latent_dim": 128,
        "prior_encoder": {
            "vocab_size": 100,
            "pitch_min": 50,
            "pitch_max": 1100,
            "pitch_bins_num": 512,
            "encoder": {
                "encoder_layer": 6,
                "encoder_hidden": 512,
                "encoder_head": 8,
                "conv_filter_size": 2048,
                "conv_kernel_size": 9,
                "encoder_dropout": 0.2,
                "use_cln": true
            },
            "duration_predictor": {
                "input_size": 512,
                "filter_size": 512,
                "kernel_size": 3,
                "conv_layers": 30,
                "cross_attn_per_layer": 3,
                "attn_head": 8,
                "drop_out": 0.5
            },
            "pitch_predictor": {
                "input_size": 512,
                "filter_size": 512,
                "kernel_size": 5,
                "conv_layers": 30,
                "cross_attn_per_layer": 3,
                "attn_head": 8,
                "drop_out": 0.5
            }
        },
        "diffusion": {
            "wavenet": {
                "input_size": 128,
                "hidden_size": 512,
                "out_size": 128,
                "num_layers": 40,
                "cross_attn_per_layer": 3,
                "dilation_cycle": 2,
                "attn_head": 8,
                "drop_out": 0.2
            },
            "beta_min": 0.05,
            "beta_max": 20,
            "sigma": 1.0,
            "noise_factor": 1.0,
            "ode_solver": "euler",
            "diffusion_type": "diffusion"
        },
        "prompt_encoder": {
            "encoder_layer": 6,
            "encoder_hidden": 512,
            "encoder_head": 8,
            "conv_filter_size": 2048,
            "conv_kernel_size": 9,
            "encoder_dropout": 0.2,
            "use_cln": false
        },
        "query_emb": {
            "query_token_num": 32,
            "hidden_size": 512,
            "head_num": 8
        },
        "inference_step": 500
    },
    "train": {
        "use_dynamic_batchsize": true,
        "max_tokens": 7500,
        "max_sentences": 32,
        "lr_warmup_steps": 5000,
        "lr_scheduler": "cosine",
        "num_train_steps": 800000,
        "adam": {
            "lr": 7.5e-5
        },
        "diff_ce_loss_lambda": 0.5,
        "diff_noise_loss_lambda": 1.0,
        "ddp": false,
        "random_seed": 114,
        "batch_size": 32,
        "epochs": 5000,
        "max_steps": 1000000,
        "total_training_steps": 800000,
        "save_summary_steps": 500,
        "save_checkpoints_steps": 2000,
        "valid_interval": 2000,
        "keep_checkpoint_max": 100
    }
}