mlx-community/IndexTTS

1{2    "model_type": "indextts",3    "bigvgan": {4        "adam_b1": 0.8,5        "adam_b2": 0.99,6        "lr_decay": 0.999998,7        "seed": 1234,8        "resblock": "1",9        "upsample_rates": [10            4,11            4,12            4,13            4,14            2,15            216        ],17        "upsample_kernel_sizes": [18            8,19            8,20            4,21            4,22            4,23            424        ],25        "upsample_initial_channel": 1536,26        "resblock_kernel_sizes": [27            3,28            7,29            1130        ],31        "resblock_dilation_sizes": [32            [33                1,34                3,35                536            ],37            [38                1,39                3,40                541            ],42            [43                1,44                3,45                546            ]47        ],48        "feat_upsample": false,49        "speaker_embedding_dim": 512,50        "cond_d_vector_in_each_upsampling_layer": true,51        "gpt_dim": 1024,52        "activation": "snakebeta",53        "snake_logscale": true,54        "use_cqtd_instead_of_mrd": true,55        "cqtd_filters": 128,56        "cqtd_max_filters": 1024,57        "cqtd_filters_scale": 1,58        "cqtd_dilations": [59            1,60            2,61            462        ],63        "cqtd_hop_lengths": [64            512,65            256,66            25667        ],68        "cqtd_n_octaves": [69            9,70            9,71            972        ],73        "cqtd_bins_per_octaves": [74            24,75            36,76            4877        ],78        "resolutions": [79            [80                1024,81                120,82                60083            ],84            [85                2048,86                240,87                120088            ],89            [90                512,91                50,92                24093            ]94        ],95        "mpd_reshapes": [96            2,97            3,98            5,99            7,100            11101        ],102        "use_spectral_norm": false,103        "discriminator_channel_mult": 1,104        "use_multiscale_melloss": true,105        "lambda_melloss": 15,106        "clip_grad_norm": 1000,107        "segment_size": 16384,108        "num_mels": 100,109        "num_freq": 1025,110        "n_fft": 1024,111        "hop_size": 256,112        "win_size": 1024,113        "sampling_rate": 24000,114        "fmin": 0,115        "fmax": null,116        "fmax_for_loss": null,117        "mel_type": "pytorch",118        "num_workers": 2,119        "dist_config": {120            "dist_backend": "nccl",121            "dist_url": "tcp://localhost:54321",122            "world_size": 1123        }124    },125    "bigvgan_checkpoint": "bigvgan_generator.pth",126    "dataset": {127        "bpe_model": "checkpoints/bpe.model",128        "sample_rate": 24000,129        "squeeze": false,130        "mel": {131            "sample_rate": 24000,132            "n_fft": 1024,133            "hop_length": 256,134            "win_length": 1024,135            "n_mels": 100,136            "mel_fmin": 0,137            "normalize": false138        }139    },140    "dvae_checkpoint": "dvae.pth",141    "gpt": {142        "model_dim": 1024,143        "max_mel_tokens": 605,144        "max_text_tokens": 402,145        "heads": 16,146        "use_mel_codes_as_input": true,147        "mel_length_compression": 1024,148        "layers": 20,149        "number_text_tokens": 12000,150        "number_mel_codes": 8194,151        "start_mel_token": 8192,152        "stop_mel_token": 8193,153        "start_text_token": 0,154        "stop_text_token": 1,155        "train_solo_embeddings": false,156        "condition_type": "conformer_perceiver",157        "condition_module": {158            "output_size": 512,159            "linear_units": 2048,160            "attention_heads": 8,161            "num_blocks": 6,162            "input_layer": "conv2d2",163            "perceiver_mult": 2164        }165    },166    "gpt_checkpoint": "gpt.pth",167    "vqvae": {168        "channels": 100,169        "num_tokens": 8192,170        "hidden_dim": 512,171        "num_resnet_blocks": 3,172        "codebook_dim": 512,173        "num_layers": 2,174        "positional_dims": 1,175        "kernel_size": 3,176        "smooth_l1_loss": true,177        "use_transposed_convs": false178    }179}