Upload 5Hz_16rvq_c4096_config.json with huggingface_hub
Browse files- 5Hz_16rvq_c4096_config.json +172 -0
5Hz_16rvq_c4096_config.json
ADDED
|
@@ -0,0 +1,172 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"trainer_type": "TokenizerGanArkTrainer",
|
| 3 |
+
"synthesizer_type": "TokenizerSynthesizer",
|
| 4 |
+
"model": {
|
| 5 |
+
"type": "TokenizerGANWrapper",
|
| 6 |
+
"tokenizer_type": "mimicodec_vq",
|
| 7 |
+
"tokenizer_conf": {
|
| 8 |
+
"seanet_conf": {
|
| 9 |
+
"channels": 1,
|
| 10 |
+
"dimension": 512,
|
| 11 |
+
"causal": false,
|
| 12 |
+
"n_filters": 64,
|
| 13 |
+
"n_residual_layers": 1,
|
| 14 |
+
"activation": "ELU",
|
| 15 |
+
"compress": 2,
|
| 16 |
+
"dilation_base": 2,
|
| 17 |
+
"disable_norm_outer_blocks": 0,
|
| 18 |
+
"kernel_size": 7,
|
| 19 |
+
"residual_kernel_size": 3,
|
| 20 |
+
"last_kernel_size": 3,
|
| 21 |
+
"norm": "none",
|
| 22 |
+
"pad_mode": "constant",
|
| 23 |
+
"ratios": [
|
| 24 |
+
8,
|
| 25 |
+
5,
|
| 26 |
+
5,
|
| 27 |
+
4,
|
| 28 |
+
4
|
| 29 |
+
],
|
| 30 |
+
"true_skip": true
|
| 31 |
+
},
|
| 32 |
+
"transformer_conf": {
|
| 33 |
+
"d_model": 512,
|
| 34 |
+
"num_heads": 8,
|
| 35 |
+
"num_layers": 8,
|
| 36 |
+
"causal": false,
|
| 37 |
+
"layer_scale": 0.01,
|
| 38 |
+
"context": 1000,
|
| 39 |
+
"conv_layout": true,
|
| 40 |
+
"max_period": 10000,
|
| 41 |
+
"gating": "none",
|
| 42 |
+
"norm": "layer_norm",
|
| 43 |
+
"positional_embedding": "rope",
|
| 44 |
+
"dim_feedforward": 2048
|
| 45 |
+
},
|
| 46 |
+
"vq_conf": {
|
| 47 |
+
"num_quantizers": 32,
|
| 48 |
+
"dim": 512,
|
| 49 |
+
"codebook_size": 256,
|
| 50 |
+
"codebook_dim": 8,
|
| 51 |
+
"threshold_ema_dead_code": 2,
|
| 52 |
+
"commitment": 0.25,
|
| 53 |
+
"weight_init": false,
|
| 54 |
+
"full_commit_loss": false
|
| 55 |
+
},
|
| 56 |
+
"ds_rate": 1,
|
| 57 |
+
"sample_rate": 16000
|
| 58 |
+
},
|
| 59 |
+
"wave_decoder_type": "mimicodec_dec",
|
| 60 |
+
"wave_decoder_conf": {
|
| 61 |
+
"in_channels": 512,
|
| 62 |
+
"seanet_conf": {
|
| 63 |
+
"channels": 1,
|
| 64 |
+
"dimension": 512,
|
| 65 |
+
"causal": false,
|
| 66 |
+
"n_filters": 64,
|
| 67 |
+
"n_residual_layers": 1,
|
| 68 |
+
"activation": "ELU",
|
| 69 |
+
"compress": 2,
|
| 70 |
+
"dilation_base": 2,
|
| 71 |
+
"disable_norm_outer_blocks": 0,
|
| 72 |
+
"kernel_size": 7,
|
| 73 |
+
"residual_kernel_size": 3,
|
| 74 |
+
"last_kernel_size": 3,
|
| 75 |
+
"norm": "none",
|
| 76 |
+
"pad_mode": "constant",
|
| 77 |
+
"ratios": [
|
| 78 |
+
8,
|
| 79 |
+
5,
|
| 80 |
+
5,
|
| 81 |
+
4,
|
| 82 |
+
4
|
| 83 |
+
],
|
| 84 |
+
"true_skip": true
|
| 85 |
+
},
|
| 86 |
+
"transformer_conf": {
|
| 87 |
+
"d_model": 512,
|
| 88 |
+
"num_heads": 8,
|
| 89 |
+
"num_layers": 8,
|
| 90 |
+
"causal": false,
|
| 91 |
+
"layer_scale": 0.01,
|
| 92 |
+
"context": 1000,
|
| 93 |
+
"conv_layout": true,
|
| 94 |
+
"max_period": 10000,
|
| 95 |
+
"gating": "none",
|
| 96 |
+
"norm": "layer_norm",
|
| 97 |
+
"positional_embedding": "rope",
|
| 98 |
+
"dim_feedforward": 2048
|
| 99 |
+
},
|
| 100 |
+
"up_rate": 1
|
| 101 |
+
},
|
| 102 |
+
"wav_input_sr": 16000,
|
| 103 |
+
"discriminators": {
|
| 104 |
+
"type": [
|
| 105 |
+
"HiFiGANMultiPeriodDiscriminator",
|
| 106 |
+
"SpecDiscriminator"
|
| 107 |
+
],
|
| 108 |
+
"HiFiGANMultiPeriodDiscriminator": {
|
| 109 |
+
"period_sizes": [
|
| 110 |
+
2,
|
| 111 |
+
3,
|
| 112 |
+
5,
|
| 113 |
+
7,
|
| 114 |
+
11
|
| 115 |
+
],
|
| 116 |
+
"max_downsample_channels": 512,
|
| 117 |
+
"channels": 16,
|
| 118 |
+
"channel_increasing_factor": 4
|
| 119 |
+
},
|
| 120 |
+
"SpecDiscriminator": {
|
| 121 |
+
"stft_params": {
|
| 122 |
+
"fft_sizes": [
|
| 123 |
+
78,
|
| 124 |
+
126,
|
| 125 |
+
206,
|
| 126 |
+
334,
|
| 127 |
+
542,
|
| 128 |
+
876,
|
| 129 |
+
1418,
|
| 130 |
+
2296
|
| 131 |
+
],
|
| 132 |
+
"hop_sizes": [
|
| 133 |
+
39,
|
| 134 |
+
63,
|
| 135 |
+
103,
|
| 136 |
+
167,
|
| 137 |
+
271,
|
| 138 |
+
438,
|
| 139 |
+
709,
|
| 140 |
+
1148
|
| 141 |
+
],
|
| 142 |
+
"win_lengths": [
|
| 143 |
+
78,
|
| 144 |
+
126,
|
| 145 |
+
206,
|
| 146 |
+
334,
|
| 147 |
+
542,
|
| 148 |
+
876,
|
| 149 |
+
1418,
|
| 150 |
+
2296
|
| 151 |
+
],
|
| 152 |
+
"window": "hann_window"
|
| 153 |
+
},
|
| 154 |
+
"in_channels": 1,
|
| 155 |
+
"out_channels": 1,
|
| 156 |
+
"kernel_sizes": [
|
| 157 |
+
5,
|
| 158 |
+
3
|
| 159 |
+
],
|
| 160 |
+
"channels": 32,
|
| 161 |
+
"max_downsample_channels": 512,
|
| 162 |
+
"downsample_scales": [
|
| 163 |
+
2,
|
| 164 |
+
2,
|
| 165 |
+
2
|
| 166 |
+
],
|
| 167 |
+
"use_weight_norm": true
|
| 168 |
+
}
|
| 169 |
+
}
|
| 170 |
+
}
|
| 171 |
+
|
| 172 |
+
}
|