Manel/X-Codec

1{2  "acoustic_model_config": {3    "codebook_dim": 8,4    "codebook_loss_weight": 1.0,5    "codebook_size": 1024,6    "commitment_loss_weight": 0.25,7    "decoder_hidden_size": 1024,8    "downsampling_ratios": [9      8,10      5,11      4,12      213    ],14    "encoder_hidden_size": 64,15    "hidden_size": 256,16    "hop_length": 320,17    "model_type": "dac",18    "n_codebooks": 9,19    "quantizer_dropout": 0,20    "sampling_rate": 16000,21    "upsampling_ratios": [22      8,23      5,24      4,25      226    ]27  },28  "audio_channels": 1,29  "block_dilations": [30    1,31    132  ],33  "channel_ratios": [34    1,35    136  ],37  "codebook_dim": 1024,38  "codebook_size": 1024,39  "decoder_channels": 768,40  "encoder_channels": 768,41  "hidden_dim": 1024,42  "initializer_range": 0.02,43  "input_channels": 768,44  "intermediate_dim": 768,45  "kernel_size": 3,46  "model_type": "xcodec",47  "num_quantizers": 8,48  "output_channels": 768,49  "output_dim": 256,50  "sample_rate": 16000,51  "semantic_model_config": {52    "activation_dropout": 0.1,53    "apply_spec_augment": true,54    "attention_dropout": 0.1,55    "classifier_proj_size": 256,56    "conv_bias": false,57    "conv_dim": [58      512,59      512,60      512,61      512,62      512,63      512,64      51265    ],66    "conv_kernel": [67      10,68      3,69      3,70      3,71      3,72      2,73      274    ],75    "conv_pos_batch_norm": false,76    "conv_stride": [77      5,78      2,79      2,80      2,81      2,82      2,83      284    ],85    "ctc_loss_reduction": "sum",86    "ctc_zero_infinity": false,87    "do_stable_layer_norm": false,88    "feat_extract_activation": "gelu",89    "feat_extract_norm": "group",90    "feat_proj_dropout": 0.0,91    "feat_proj_layer_norm": true,92    "final_dropout": 0.1,93    "hidden_act": "gelu",94    "hidden_dropout": 0.1,95    "hidden_size": 768,96    "initializer_range": 0.02,97    "intermediate_size": 3072,98    "layer_norm_eps": 1e-05,99    "layerdrop": 0.1,100    "mask_feature_length": 10,101    "mask_feature_min_masks": 0,102    "mask_feature_prob": 0.0,103    "mask_time_length": 10,104    "mask_time_min_masks": 2,105    "mask_time_prob": 0.05,106    "model_type": "hubert",107    "num_attention_heads": 12,108    "num_conv_pos_embedding_groups": 16,109    "num_conv_pos_embeddings": 128,110    "num_feat_extract_layers": 7,111    "num_hidden_layers": 12,112    "use_weighted_layer_sum": false,113    "vocab_size": 32114  },115  "strides": [116    1,117    1118  ],119  "target_bandwidths": [120    0.5,121    1,122    1.5,123    2,124    4125  ],126  "transformers_version": "4.52.0.dev0",127  "unit_kernel_size": 3128}129