audio_tokenizer/config.json
| 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 | 2, |
| 13 | 3 |
| 14 | ], |
| 15 | "encoder_hidden_size": 64, |
| 16 | "hidden_size": 256, |
| 17 | "hop_length": 960, |
| 18 | "model_type": "dac", |
| 19 | "n_codebooks": 9, |
| 20 | "quantizer_dropout": 0, |
| 21 | "sampling_rate": 16000, |
| 22 | "upsampling_ratios": [ |
| 23 | 8, |
| 24 | 5, |
| 25 | 4, |
| 26 | 2, |
| 27 | 3 |
| 28 | ] |
| 29 | }, |
| 30 | "architectures": [ |
| 31 | "HiggsAudioV2TokenizerModel" |
| 32 | ], |
| 33 | "block_dilations": [ |
| 34 | 1, |
| 35 | 1 |
| 36 | ], |
| 37 | "channel_ratios": [ |
| 38 | 1, |
| 39 | 1 |
| 40 | ], |
| 41 | "codebook_dim": 64, |
| 42 | "codebook_size": 1024, |
| 43 | "downsample_factor": 320, |
| 44 | "dtype": "float32", |
| 45 | "initializer_range": 0.02, |
| 46 | "kernel_size": 3, |
| 47 | "model_type": "higgs_audio_v2_tokenizer", |
| 48 | "sample_rate": 24000, |
| 49 | "semantic_model_config": { |
| 50 | "activation_dropout": 0.1, |
| 51 | "apply_spec_augment": true, |
| 52 | "attention_dropout": 0.1, |
| 53 | "bos_token_id": 1, |
| 54 | "classifier_proj_size": 256, |
| 55 | "conv_bias": false, |
| 56 | "conv_dim": [ |
| 57 | 512, |
| 58 | 512, |
| 59 | 512, |
| 60 | 512, |
| 61 | 512, |
| 62 | 512, |
| 63 | 512 |
| 64 | ], |
| 65 | "conv_kernel": [ |
| 66 | 10, |
| 67 | 3, |
| 68 | 3, |
| 69 | 3, |
| 70 | 3, |
| 71 | 2, |
| 72 | 2 |
| 73 | ], |
| 74 | "conv_pos_batch_norm": false, |
| 75 | "conv_stride": [ |
| 76 | 5, |
| 77 | 2, |
| 78 | 2, |
| 79 | 2, |
| 80 | 2, |
| 81 | 2, |
| 82 | 2 |
| 83 | ], |
| 84 | "ctc_loss_reduction": "sum", |
| 85 | "ctc_zero_infinity": false, |
| 86 | "do_stable_layer_norm": false, |
| 87 | "eos_token_id": 2, |
| 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": 0.0, |
| 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 | "pad_token_id": 0, |
| 113 | "use_weighted_layer_sum": false, |
| 114 | "vocab_size": 32 |
| 115 | }, |
| 116 | "semantic_sample_rate": 16000, |
| 117 | "strides": [ |
| 118 | 1, |
| 119 | 1 |
| 120 | ], |
| 121 | "target_bandwidths": [ |
| 122 | 0.5, |
| 123 | 1, |
| 124 | 1.5, |
| 125 | 2 |
| 126 | ], |
| 127 | "transformers_version": "5.3.0.dev0", |
| 128 | "unit_kernel_size": 3 |
| 129 | } |
| 130 | |