speech_tokenizer/config.json
| 1 | { |
| 2 | "architectures": [ |
| 3 | "Qwen3TTSTokenizerV2Model" |
| 4 | ], |
| 5 | "model_type": "qwen3_tts_tokenizer_12hz", |
| 6 | "encoder_valid_num_quantizers": 16, |
| 7 | "input_sample_rate": 24000, |
| 8 | "output_sample_rate": 24000, |
| 9 | "decode_upsample_rate": 1920, |
| 10 | "encode_downsample_rate": 1920, |
| 11 | "decoder_config": { |
| 12 | "attention_bias": false, |
| 13 | "attention_dropout": 0.0, |
| 14 | "latent_dim": 1024, |
| 15 | "codebook_dim": 512, |
| 16 | "codebook_size": 2048, |
| 17 | "decoder_dim": 1536, |
| 18 | "hidden_act": "silu", |
| 19 | "hidden_size": 512, |
| 20 | "intermediate_size": 1024, |
| 21 | "layer_scale_initial_scale": 0.01, |
| 22 | "max_position_embeddings": 8000, |
| 23 | "head_dim": 64, |
| 24 | "num_attention_heads": 16, |
| 25 | "num_hidden_layers": 8, |
| 26 | "num_key_value_heads": 16, |
| 27 | "num_quantizers": 16, |
| 28 | "num_semantic_quantizers": 1, |
| 29 | "rms_norm_eps": 1e-05, |
| 30 | "rope_theta": 10000, |
| 31 | "semantic_codebook_size": 4096, |
| 32 | "sliding_window": 72, |
| 33 | "upsample_rates": [ |
| 34 | 8, |
| 35 | 5, |
| 36 | 4, |
| 37 | 3 |
| 38 | ], |
| 39 | "upsampling_ratios": [ |
| 40 | 2, |
| 41 | 2 |
| 42 | ], |
| 43 | "vector_quantization_hidden_dimension": 512 |
| 44 | }, |
| 45 | "encoder_config": { |
| 46 | "_frame_rate": 12.5, |
| 47 | "attention_bias": false, |
| 48 | "attention_dropout": 0.0, |
| 49 | "audio_channels": 1, |
| 50 | "codebook_dim": 256, |
| 51 | "codebook_size": 2048, |
| 52 | "compress": 2, |
| 53 | "dilation_growth_rate": 2, |
| 54 | "dtype": "float32", |
| 55 | "head_dim": 64, |
| 56 | "hidden_act": "gelu", |
| 57 | "hidden_size": 512, |
| 58 | "initializer_range": 0.02, |
| 59 | "intermediate_size": 2048, |
| 60 | "kernel_size": 7, |
| 61 | "last_kernel_size": 3, |
| 62 | "layer_scale_initial_scale": 0.01, |
| 63 | "max_position_embeddings": 8000, |
| 64 | "norm_eps": 1e-05, |
| 65 | "normalize": false, |
| 66 | "num_attention_heads": 8, |
| 67 | "num_filters": 64, |
| 68 | "num_hidden_layers": 8, |
| 69 | "num_key_value_heads": 8, |
| 70 | "num_quantizers": 32, |
| 71 | "num_residual_layers": 1, |
| 72 | "num_semantic_quantizers": 1, |
| 73 | "pad_mode": "constant", |
| 74 | "residual_kernel_size": 3, |
| 75 | "rope_theta": 10000.0, |
| 76 | "sampling_rate": 24000, |
| 77 | "sliding_window": 250, |
| 78 | "transformers_version": "4.57.0.dev0", |
| 79 | "trim_right_ratio": 1.0, |
| 80 | "upsample_groups": 512, |
| 81 | "upsampling_ratios": [ |
| 82 | 8, |
| 83 | 6, |
| 84 | 5, |
| 85 | 4 |
| 86 | ], |
| 87 | "use_cache": false, |
| 88 | "use_causal_conv": true, |
| 89 | "use_conv_shortcut": false, |
| 90 | "use_streaming": false, |
| 91 | "vector_quantization_hidden_dimension": 256 |
| 92 | }, |
| 93 | "transformers_version": "4.57.3" |
| 94 | } |
| 95 | |