ICTNLP/indextts2_decoder

1semantic_codec:2    conf:3        codebook_size: 81924        hidden_size: 10245        codebook_dim: 86        vocos_dim: 3847        vocos_intermediate_dim: 20488        vocos_num_layers: 129    checkpoint: semantic_codec.safetensors10 11s2mel:12    preprocess_params:13        sr: 2205014        spect_params:15            n_fft: 102416            win_length: 102417            hop_length: 25618            n_mels: 8019            fmin: 020            fmax: "None"21 22    dit_type: "DiT"23    reg_loss_type: "l1"24    style_encoder:25        dim: 19226    length_regulator:27        channels: 51228        is_discrete: false29        in_channels: 102430        content_codebook_size: 204831        sampling_ratios: [1, 1, 1, 1]32        vector_quantize: false33        n_codebooks: 134        quantizer_dropout: 0.035        f0_condition: false36        n_f0_bins: 51237    DiT:38        hidden_dim: 51239        num_heads: 840        depth: 1341        class_dropout_prob: 0.142        block_size: 819243        in_channels: 8044        style_condition: true45        final_layer_type: 'wavenet'46        target: 'mel'47        content_dim: 51248        content_codebook_size: 102449        content_type: 'discrete'50        f0_condition: false51        n_f0_bins: 51252        content_codebooks: 153        is_causal: false54        long_skip_connection: true55        zero_prompt_speech_token: false56        time_as_token: false57        style_as_token: false58        uvit_skip_connection: true59        add_resblock_in_transformer: false60    wavenet:61        hidden_dim: 51262        num_layers: 863        kernel_size: 564        dilation_rate: 165        p_dropout: 0.266        style_condition: true67 68gpt_checkpoint: gpt.pth69s2mel_checkpoint: s2mel.pth70vocoder:71    type: "bigvgan"72    name: "nvidia/bigvgan_v2_22khz_80band_256x"