Upload config.json
Browse files- config.json +122 -0
config.json
ADDED
@@ -0,0 +1,122 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_type": "autoencoder",
|
3 |
+
"sample_size": 16000,
|
4 |
+
"sample_rate": 16000,
|
5 |
+
"audio_channels": 1,
|
6 |
+
"model": {
|
7 |
+
"encoder": {
|
8 |
+
"type": "oobleck",
|
9 |
+
"config": {
|
10 |
+
"in_channels": 1,
|
11 |
+
"channels": 128,
|
12 |
+
"c_mults": [1, 2, 4, 8],
|
13 |
+
"strides": [2, 4, 4, 10],
|
14 |
+
"latent_dim": 256,
|
15 |
+
"use_snake": true
|
16 |
+
}
|
17 |
+
},
|
18 |
+
"decoder": {
|
19 |
+
"type": "oobleck",
|
20 |
+
"config": {
|
21 |
+
"out_channels": 1,
|
22 |
+
"channels": 128,
|
23 |
+
"c_mults": [1, 2, 4, 8],
|
24 |
+
"strides": [2, 4, 4, 10],
|
25 |
+
"latent_dim": 128,
|
26 |
+
"use_snake": true,
|
27 |
+
"final_tanh": false
|
28 |
+
}
|
29 |
+
},
|
30 |
+
"bottleneck": {
|
31 |
+
"type": "vae"
|
32 |
+
},
|
33 |
+
"latent_dim": 128,
|
34 |
+
"downsampling_ratio": 320,
|
35 |
+
"io_channels": 1
|
36 |
+
},
|
37 |
+
"training": {
|
38 |
+
"learning_rate": 1.5e-4,
|
39 |
+
"warmup_steps": 0,
|
40 |
+
"use_ema": false,
|
41 |
+
"optimizer_configs": {
|
42 |
+
"autoencoder": {
|
43 |
+
"optimizer": {
|
44 |
+
"type": "AdamW",
|
45 |
+
"config": {
|
46 |
+
"betas": [0.8, 0.99],
|
47 |
+
"lr": 1.5e-4,
|
48 |
+
"weight_decay": 1e-3
|
49 |
+
}
|
50 |
+
},
|
51 |
+
"scheduler": {
|
52 |
+
"type": "InverseLR",
|
53 |
+
"config": {
|
54 |
+
"inv_gamma": 200000,
|
55 |
+
"power": 0.5,
|
56 |
+
"warmup": 0.999
|
57 |
+
}
|
58 |
+
}
|
59 |
+
},
|
60 |
+
"discriminator": {
|
61 |
+
"optimizer": {
|
62 |
+
"type": "AdamW",
|
63 |
+
"config": {
|
64 |
+
"betas": [0.8, 0.99],
|
65 |
+
"lr": 3e-4,
|
66 |
+
"weight_decay": 1e-3
|
67 |
+
}
|
68 |
+
},
|
69 |
+
"scheduler": {
|
70 |
+
"type": "InverseLR",
|
71 |
+
"config": {
|
72 |
+
"inv_gamma": 200000,
|
73 |
+
"power": 0.5,
|
74 |
+
"warmup": 0.999
|
75 |
+
}
|
76 |
+
}
|
77 |
+
}
|
78 |
+
},
|
79 |
+
"loss_configs": {
|
80 |
+
"discriminator": {
|
81 |
+
"type": "encodec",
|
82 |
+
"config": {
|
83 |
+
"filters": 64,
|
84 |
+
"n_ffts": [1280, 640, 320, 160, 80],
|
85 |
+
"hop_lengths": [320, 160, 80, 40, 20],
|
86 |
+
"win_lengths": [1280, 640, 320, 160, 80]
|
87 |
+
},
|
88 |
+
"weights": {
|
89 |
+
"adversarial": 0.1,
|
90 |
+
"feature_matching": 5.0
|
91 |
+
}
|
92 |
+
},
|
93 |
+
"spectral": {
|
94 |
+
"type": "mrstft",
|
95 |
+
"config": {
|
96 |
+
"fft_sizes": [1280, 640, 320, 160, 80, 40, 20],
|
97 |
+
"hop_sizes": [320, 160, 80, 40, 20, 10, 5],
|
98 |
+
"win_lengths": [1280, 640, 320, 160, 80, 40, 20],
|
99 |
+
"perceptual_weighting": true
|
100 |
+
},
|
101 |
+
"weights": {
|
102 |
+
"mrstft": 1.0
|
103 |
+
}
|
104 |
+
},
|
105 |
+
"time": {
|
106 |
+
"type": "l1",
|
107 |
+
"weights": {
|
108 |
+
"l1": 0.0
|
109 |
+
}
|
110 |
+
},
|
111 |
+
"bottleneck": {
|
112 |
+
"type": "kl",
|
113 |
+
"weights": {
|
114 |
+
"kl": 1e-4
|
115 |
+
}
|
116 |
+
}
|
117 |
+
},
|
118 |
+
"demo": {
|
119 |
+
"demo_every": 10000
|
120 |
+
}
|
121 |
+
}
|
122 |
+
}
|