Hugging Face's logo

zhisheng01
/
CtrlSpeech

Text-to-Speech
ONNX
Safetensors
English
ctrlspeech
speech-synthesis
prosody-control
controllable-tts
flow-matching
Model card Files
xet
 Community
zhisheng01 commited on
Commit
5d8e41e
·
verified ·
1 Parent(s): 9e21b96

Add CtrlSpeech checkpoints, SVAE vocoder and support files

Browse files
Files changed (18) hide show
  1. base-150m/config.yaml +49 -0
  2. base-150m/model.safetensors +3 -0
  3. base-150m/qwen_config.json +30 -0
  4. base-600m/config.yaml +49 -0
  5. base-600m/model.safetensors +3 -0
  6. base-600m/qwen_config.json +30 -0
  7. control-150m/config.yaml +49 -0
  8. control-150m/model.safetensors +3 -0
  9. control-150m/qwen_config.json +30 -0
  10. control-600m/config.yaml +49 -0
  11. control-600m/model.safetensors +3 -0
  12. control-600m/qwen_config.json +30 -0
  13. shared/campplus.onnx +3 -0
  14. shared/vocab.json +44 -0
  15. svae/config.json +1066 -0
  16. svae/dac/ema_state_dict.pth +3 -0
  17. svae/dac/weights.pth +3 -0
  18. svae/metainfo.json +1 -0
base-150m/config.yaml ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ name: DiTAR
3
+ text_vocab_size: 151643
4
+ audio_type: vae_online
5
+ audio_channels: 64
6
+ patch_size: 4
7
+ dim: 512
8
+ mlp_hidden_dim: 512
9
+ use_seperate_linear: true
10
+ aggregation_encoder:
11
+ name: qwen
12
+ hidden_size: 512
13
+ intermediate_size: 2048
14
+ num_attention_heads: 8
15
+ num_hidden_layers: 4
16
+ patch_size: 4
17
+ pool_type: cls
18
+ backbone:
19
+ name: Qwen
20
+ version: qwen3
21
+ qwen_config_path: qwen_config.json
22
+ pretrained_LM_path: null
23
+ freeze_ar: false
24
+ load_pretrained_weights: false
25
+ load_phoneme: true
26
+ weighted_layers: false
27
+ loc_decoder:
28
+ name: DiT
29
+ model:
30
+ dim: 512
31
+ out_dim: 64
32
+ depth: 4
33
+ heads: 8
34
+ ff_mult: 4
35
+ qk_norm: null
36
+ pe_attn_head: null
37
+ checkpoint_activations: false
38
+ random_time: false
39
+ time_schedule: false
40
+ history_vae_window_size: 4
41
+ drop_cond_prob: 0.1
42
+ drop_condition: drop_ctx_or_his
43
+ stop_predictor_type: transformer
44
+ vocoder:
45
+ path: ../svae
46
+ loss:
47
+ use_ar_l1_loss: true
48
+ use_stop_loss: true
49
+ use_vae_projected_l1_loss: true
base-150m/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec3bced46c6047804145dc2a3ea93a8213d05ce1b6498ab07fdac848dd59009b
3
+ size 600532788
base-150m/qwen_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 151643,
8
+ "eos_token_id": 151645,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 512,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 1024,
14
+ "max_position_embeddings": 40960,
15
+ "max_window_layers": 28,
16
+ "model_type": "qwen3",
17
+ "num_attention_heads": 8,
18
+ "num_hidden_layers": 24,
19
+ "num_key_value_heads": 8,
20
+ "rms_norm_eps": 1e-06,
21
+ "rope_scaling": null,
22
+ "rope_theta": 1000000,
23
+ "sliding_window": null,
24
+ "tie_word_embeddings": true,
25
+ "torch_dtype": "bfloat16",
26
+ "transformers_version": "4.51.0",
27
+ "use_cache": true,
28
+ "use_sliding_window": false,
29
+ "vocab_size": 151936
30
+ }
base-600m/config.yaml ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ name: DiTAR
3
+ text_vocab_size: 151643
4
+ audio_type: vae_online
5
+ audio_channels: 64
6
+ patch_size: 4
7
+ dim: 1024
8
+ mlp_hidden_dim: 512
9
+ use_seperate_linear: true
10
+ aggregation_encoder:
11
+ name: qwen
12
+ hidden_size: 1024
13
+ intermediate_size: 4096
14
+ num_attention_heads: 16
15
+ num_hidden_layers: 6
16
+ patch_size: 4
17
+ pool_type: cls
18
+ backbone:
19
+ name: Qwen
20
+ version: qwen3
21
+ qwen_config_path: qwen_config.json
22
+ pretrained_LM_path: null
23
+ freeze_ar: false
24
+ load_pretrained_weights: false
25
+ load_phoneme: true
26
+ weighted_layers: false
27
+ loc_decoder:
28
+ name: DiT
29
+ model:
30
+ dim: 1024
31
+ out_dim: 64
32
+ depth: 6
33
+ heads: 16
34
+ ff_mult: 4
35
+ qk_norm: null
36
+ pe_attn_head: null
37
+ checkpoint_activations: false
38
+ random_time: false
39
+ time_schedule: false
40
+ history_vae_window_size: 4
41
+ drop_cond_prob: 0.1
42
+ drop_condition: drop_ctx_or_his
43
+ stop_predictor_type: transformer
44
+ vocoder:
45
+ path: ../svae
46
+ loss:
47
+ use_ar_l1_loss: true
48
+ use_stop_loss: true
49
+ use_vae_projected_l1_loss: true
base-600m/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fedb2ba873b81bc9b03794c926d7846dbfb5ea98bd5e85ad2c6233f57d89695
3
+ size 2767307788
base-600m/qwen_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 151643,
8
+ "eos_token_id": 151645,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 1024,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 3072,
14
+ "max_position_embeddings": 40960,
15
+ "max_window_layers": 28,
16
+ "model_type": "qwen3",
17
+ "num_attention_heads": 16,
18
+ "num_hidden_layers": 28,
19
+ "num_key_value_heads": 8,
20
+ "rms_norm_eps": 1e-06,
21
+ "rope_scaling": null,
22
+ "rope_theta": 1000000,
23
+ "sliding_window": null,
24
+ "tie_word_embeddings": true,
25
+ "torch_dtype": "bfloat16",
26
+ "transformers_version": "4.51.0",
27
+ "use_cache": true,
28
+ "use_sliding_window": false,
29
+ "vocab_size": 151936
30
+ }
control-150m/config.yaml ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ name: DiTAR
3
+ text_vocab_size: 151643
4
+ audio_type: vae_online
5
+ audio_channels: 64
6
+ patch_size: 4
7
+ dim: 512
8
+ mlp_hidden_dim: 512
9
+ use_seperate_linear: true
10
+ aggregation_encoder:
11
+ name: qwen
12
+ hidden_size: 512
13
+ intermediate_size: 2048
14
+ num_attention_heads: 8
15
+ num_hidden_layers: 4
16
+ patch_size: 4
17
+ pool_type: cls
18
+ backbone:
19
+ name: Qwen
20
+ version: qwen3
21
+ qwen_config_path: qwen_config.json
22
+ pretrained_LM_path: null
23
+ freeze_ar: false
24
+ load_pretrained_weights: false
25
+ load_phoneme: true
26
+ weighted_layers: false
27
+ loc_decoder:
28
+ name: DiT
29
+ model:
30
+ dim: 512
31
+ out_dim: 64
32
+ depth: 4
33
+ heads: 8
34
+ ff_mult: 4
35
+ qk_norm: null
36
+ pe_attn_head: null
37
+ checkpoint_activations: false
38
+ random_time: false
39
+ time_schedule: false
40
+ history_vae_window_size: 4
41
+ drop_cond_prob: 0.1
42
+ drop_condition: drop_ctx_or_his
43
+ stop_predictor_type: transformer
44
+ vocoder:
45
+ path: ../svae
46
+ loss:
47
+ use_ar_l1_loss: true
48
+ use_stop_loss: true
49
+ use_vae_projected_l1_loss: true
control-150m/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25810d7566a3915678f0685e96a3aae8c8e3ba2826b3503f8c8a75e0d937df6d
3
+ size 601319532
control-150m/qwen_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 151643,
8
+ "eos_token_id": 151645,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 512,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 1024,
14
+ "max_position_embeddings": 40960,
15
+ "max_window_layers": 28,
16
+ "model_type": "qwen3",
17
+ "num_attention_heads": 8,
18
+ "num_hidden_layers": 24,
19
+ "num_key_value_heads": 8,
20
+ "rms_norm_eps": 1e-06,
21
+ "rope_scaling": null,
22
+ "rope_theta": 1000000,
23
+ "sliding_window": null,
24
+ "tie_word_embeddings": true,
25
+ "torch_dtype": "bfloat16",
26
+ "transformers_version": "4.51.0",
27
+ "use_cache": true,
28
+ "use_sliding_window": false,
29
+ "vocab_size": 151936
30
+ }
control-600m/config.yaml ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ name: DiTAR
3
+ text_vocab_size: 151643
4
+ audio_type: vae_online
5
+ audio_channels: 64
6
+ patch_size: 4
7
+ dim: 1024
8
+ mlp_hidden_dim: 512
9
+ use_seperate_linear: true
10
+ aggregation_encoder:
11
+ name: qwen
12
+ hidden_size: 1024
13
+ intermediate_size: 4096
14
+ num_attention_heads: 16
15
+ num_hidden_layers: 6
16
+ patch_size: 4
17
+ pool_type: cls
18
+ backbone:
19
+ name: Qwen
20
+ version: qwen3
21
+ qwen_config_path: qwen_config.json
22
+ pretrained_LM_path: null
23
+ freeze_ar: false
24
+ load_pretrained_weights: false
25
+ load_phoneme: true
26
+ weighted_layers: false
27
+ loc_decoder:
28
+ name: DiT
29
+ model:
30
+ dim: 1024
31
+ out_dim: 64
32
+ depth: 6
33
+ heads: 16
34
+ ff_mult: 4
35
+ qk_norm: null
36
+ pe_attn_head: null
37
+ checkpoint_activations: false
38
+ random_time: false
39
+ time_schedule: false
40
+ history_vae_window_size: 4
41
+ drop_cond_prob: 0.1
42
+ drop_condition: drop_ctx_or_his
43
+ stop_predictor_type: transformer
44
+ vocoder:
45
+ path: ../svae
46
+ loss:
47
+ use_ar_l1_loss: true
48
+ use_stop_loss: true
49
+ use_vae_projected_l1_loss: true
control-600m/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d30764c44c099bbe561d05e2932eae16b512649ed803c831a808d72157d9007
3
+ size 2768880972
control-600m/qwen_config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 151643,
8
+ "eos_token_id": 151645,
9
+ "head_dim": 128,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 1024,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 3072,
14
+ "max_position_embeddings": 40960,
15
+ "max_window_layers": 28,
16
+ "model_type": "qwen3",
17
+ "num_attention_heads": 16,
18
+ "num_hidden_layers": 28,
19
+ "num_key_value_heads": 8,
20
+ "rms_norm_eps": 1e-06,
21
+ "rope_scaling": null,
22
+ "rope_theta": 1000000,
23
+ "sliding_window": null,
24
+ "tie_word_embeddings": true,
25
+ "torch_dtype": "bfloat16",
26
+ "transformers_version": "4.51.0",
27
+ "use_cache": true,
28
+ "use_sliding_window": false,
29
+ "vocab_size": 151936
30
+ }
shared/campplus.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6ac6a63997761ae2997373e2ee1c47040854b4b759ea41ec48e4e42df0f4d73
3
+ size 28303423
shared/vocab.json ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "[PAD]": 0,
3
+ "[UNK]": 1,
4
+ "AA": 2,
5
+ "AE": 3,
6
+ "AH": 4,
7
+ "AO": 5,
8
+ "AW": 6,
9
+ "AY": 7,
10
+ "B": 8,
11
+ "CH": 9,
12
+ "D": 10,
13
+ "DH": 11,
14
+ "EH": 12,
15
+ "ER": 13,
16
+ "EY": 14,
17
+ "F": 15,
18
+ "G": 16,
19
+ "HH": 17,
20
+ "IH": 18,
21
+ "IY": 19,
22
+ "JH": 20,
23
+ "K": 21,
24
+ "L": 22,
25
+ "M": 23,
26
+ "N": 24,
27
+ "NG": 25,
28
+ "OW": 26,
29
+ "OY": 27,
30
+ "P": 28,
31
+ "R": 29,
32
+ "S": 30,
33
+ "SH": 31,
34
+ "T": 32,
35
+ "TH": 33,
36
+ "UH": 34,
37
+ "UW": 35,
38
+ "V": 36,
39
+ "W": 37,
40
+ "Y": 38,
41
+ "Z": 39,
42
+ "ZH": 40,
43
+ "|": 41
44
+ }
svae/config.json ADDED
@@ -0,0 +1,1066 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "AdamW.betas": [
3
+ 0.8,
4
+ 0.99
5
+ ],
6
+ "generator/AdamW.betas": [
7
+ 0.8,
8
+ 0.99
9
+ ],
10
+ "discriminator/AdamW.betas": [
11
+ 0.8,
12
+ 0.99
13
+ ],
14
+ "AdamW.eps": 1e-08,
15
+ "generator/AdamW.eps": 1e-08,
16
+ "discriminator/AdamW.eps": 1e-08,
17
+ "AdamW.weight_decay": 0.01,
18
+ "generator/AdamW.weight_decay": 0.01,
19
+ "discriminator/AdamW.weight_decay": 0.01,
20
+ "AdamW.amsgrad": false,
21
+ "generator/AdamW.amsgrad": false,
22
+ "discriminator/AdamW.amsgrad": false,
23
+ "AdamW.maximize": false,
24
+ "generator/AdamW.maximize": false,
25
+ "discriminator/AdamW.maximize": false,
26
+ "AdamW.capturable": false,
27
+ "generator/AdamW.capturable": false,
28
+ "discriminator/AdamW.capturable": false,
29
+ "AdamW.differentiable": false,
30
+ "generator/AdamW.differentiable": false,
31
+ "discriminator/AdamW.differentiable": false,
32
+ "amp": false,
33
+ "ExponentialLR.gamma": 0.999996,
34
+ "generator/ExponentialLR.gamma": 0.999996,
35
+ "discriminator/ExponentialLR.gamma": 0.999996,
36
+ "DAC.encoder_dim": 64,
37
+ "DAC.encoder_rates": [
38
+ 4,
39
+ 4,
40
+ 5,
41
+ 5
42
+ ],
43
+ "DAC.latent_dim": null,
44
+ "DAC.decoder_dim": 1024,
45
+ "DAC.decoder_rates": [
46
+ 5,
47
+ 5,
48
+ 4,
49
+ 4
50
+ ],
51
+ "DAC.sample_rate": 16000,
52
+ "DAC.distill": true,
53
+ "DAC.distill_hidden_dim": 1024,
54
+ "DAC.decoder_type": "bigvgan",
55
+ "DAC.pre_vae_block": false,
56
+ "DAC.attn_proj": true,
57
+ "DAC.post_vae_block": false,
58
+ "DAC.bigvgan_conf": "conf/bigvgan_conf/bigvgan_v2_16khz_40hz_base_vae64.json",
59
+ "DAC.align_space": "vae",
60
+ "DAC.sampling_ratios": [
61
+ 0,
62
+ 1
63
+ ],
64
+ "Discriminator.rates": [],
65
+ "Discriminator.periods": [
66
+ 2,
67
+ 3,
68
+ 5,
69
+ 7,
70
+ 11
71
+ ],
72
+ "Discriminator.fft_sizes": [
73
+ 2048,
74
+ 1024,
75
+ 512
76
+ ],
77
+ "Discriminator.sample_rate": 16000,
78
+ "Discriminator.bands": [
79
+ [
80
+ 0.0,
81
+ 0.1
82
+ ],
83
+ [
84
+ 0.1,
85
+ 0.25
86
+ ],
87
+ [
88
+ 0.25,
89
+ 0.5
90
+ ],
91
+ [
92
+ 0.5,
93
+ 0.75
94
+ ],
95
+ [
96
+ 0.75,
97
+ 1.0
98
+ ]
99
+ ],
100
+ "Discriminator.use_msstft_replace_mrd": false,
101
+ "AudioDataset.n_examples": 1000,
102
+ "train/AudioDataset.n_examples": 10000000,
103
+ "val/AudioDataset.n_examples": 250,
104
+ "AudioDataset.duration": 0.5,
105
+ "train/AudioDataset.duration": 3.0,
106
+ "val/AudioDataset.duration": 5.0,
107
+ "AudioDataset.offset": null,
108
+ "train/AudioDataset.offset": null,
109
+ "val/AudioDataset.offset": null,
110
+ "AudioDataset.loudness_cutoff": -40,
111
+ "train/AudioDataset.loudness_cutoff": -40,
112
+ "val/AudioDataset.loudness_cutoff": -40,
113
+ "AudioDataset.num_channels": 1,
114
+ "train/AudioDataset.num_channels": 1,
115
+ "val/AudioDataset.num_channels": 1,
116
+ "AudioDataset.aligned": false,
117
+ "train/AudioDataset.aligned": false,
118
+ "val/AudioDataset.aligned": false,
119
+ "AudioDataset.shuffle_loaders": false,
120
+ "train/AudioDataset.shuffle_loaders": false,
121
+ "val/AudioDataset.shuffle_loaders": false,
122
+ "AudioDataset.without_replacement": true,
123
+ "train/AudioDataset.without_replacement": true,
124
+ "val/AudioDataset.without_replacement": true,
125
+ "AudioLoader.sources": null,
126
+ "train/AudioLoader.sources": null,
127
+ "val/AudioLoader.sources": null,
128
+ "AudioLoader.weights": null,
129
+ "train/AudioLoader.weights": null,
130
+ "val/AudioLoader.weights": null,
131
+ "AudioLoader.relative_path": "",
132
+ "train/AudioLoader.relative_path": "",
133
+ "val/AudioLoader.relative_path": "",
134
+ "AudioLoader.ext": [
135
+ ".wav",
136
+ ".flac",
137
+ ".mp3",
138
+ ".mp4"
139
+ ],
140
+ "train/AudioLoader.ext": [
141
+ ".wav",
142
+ ".flac",
143
+ ".mp3",
144
+ ".mp4"
145
+ ],
146
+ "val/AudioLoader.ext": [
147
+ ".wav",
148
+ ".flac",
149
+ ".mp3",
150
+ ".mp4"
151
+ ],
152
+ "AudioLoader.shuffle": true,
153
+ "train/AudioLoader.shuffle": true,
154
+ "val/AudioLoader.shuffle": true,
155
+ "AudioLoader.shuffle_state": 0,
156
+ "train/AudioLoader.shuffle_state": 0,
157
+ "val/AudioLoader.shuffle_state": 0,
158
+ "AudioLoader.guidance_frame_rate": 50,
159
+ "train/AudioLoader.guidance_frame_rate": 50,
160
+ "val/AudioLoader.guidance_frame_rate": 50,
161
+ "BackgroundNoise.snr": [
162
+ "uniform",
163
+ 10.0,
164
+ 30.0
165
+ ],
166
+ "train/BackgroundNoise.snr": [
167
+ "uniform",
168
+ 10.0,
169
+ 30.0
170
+ ],
171
+ "val/BackgroundNoise.snr": [
172
+ "uniform",
173
+ 10.0,
174
+ 30.0
175
+ ],
176
+ "BackgroundNoise.sources": null,
177
+ "train/BackgroundNoise.sources": null,
178
+ "val/BackgroundNoise.sources": null,
179
+ "BackgroundNoise.weights": null,
180
+ "train/BackgroundNoise.weights": null,
181
+ "val/BackgroundNoise.weights": null,
182
+ "BackgroundNoise.eq_amount": [
183
+ "const",
184
+ 1.0
185
+ ],
186
+ "train/BackgroundNoise.eq_amount": [
187
+ "const",
188
+ 1.0
189
+ ],
190
+ "val/BackgroundNoise.eq_amount": [
191
+ "const",
192
+ 1.0
193
+ ],
194
+ "BackgroundNoise.n_bands": 3,
195
+ "train/BackgroundNoise.n_bands": 3,
196
+ "val/BackgroundNoise.n_bands": 3,
197
+ "BackgroundNoise.name": null,
198
+ "train/BackgroundNoise.name": null,
199
+ "val/BackgroundNoise.name": null,
200
+ "BackgroundNoise.prob": 1.0,
201
+ "train/BackgroundNoise.prob": 1.0,
202
+ "val/BackgroundNoise.prob": 1.0,
203
+ "BackgroundNoise.loudness_cutoff": null,
204
+ "train/BackgroundNoise.loudness_cutoff": null,
205
+ "val/BackgroundNoise.loudness_cutoff": null,
206
+ "ClippingDistortion.perc": [
207
+ "uniform",
208
+ 0.0,
209
+ 0.1
210
+ ],
211
+ "train/ClippingDistortion.perc": [
212
+ "uniform",
213
+ 0.0,
214
+ 0.1
215
+ ],
216
+ "val/ClippingDistortion.perc": [
217
+ "uniform",
218
+ 0.0,
219
+ 0.1
220
+ ],
221
+ "ClippingDistortion.name": null,
222
+ "train/ClippingDistortion.name": null,
223
+ "val/ClippingDistortion.name": null,
224
+ "ClippingDistortion.prob": 1.0,
225
+ "train/ClippingDistortion.prob": 1.0,
226
+ "val/ClippingDistortion.prob": 1.0,
227
+ "CorruptPhase.scale": [
228
+ "uniform",
229
+ 0,
230
+ 3.141592653589793
231
+ ],
232
+ "train/CorruptPhase.scale": [
233
+ "uniform",
234
+ 0,
235
+ 3.141592653589793
236
+ ],
237
+ "val/CorruptPhase.scale": [
238
+ "uniform",
239
+ 0,
240
+ 3.141592653589793
241
+ ],
242
+ "CorruptPhase.name": null,
243
+ "train/CorruptPhase.name": null,
244
+ "val/CorruptPhase.name": null,
245
+ "CorruptPhase.prob": 1,
246
+ "train/CorruptPhase.prob": 1,
247
+ "val/CorruptPhase.prob": 1,
248
+ "CrossTalk.snr": [
249
+ "uniform",
250
+ 0.0,
251
+ 10.0
252
+ ],
253
+ "train/CrossTalk.snr": [
254
+ "uniform",
255
+ 0.0,
256
+ 10.0
257
+ ],
258
+ "val/CrossTalk.snr": [
259
+ "uniform",
260
+ 0.0,
261
+ 10.0
262
+ ],
263
+ "CrossTalk.sources": null,
264
+ "train/CrossTalk.sources": null,
265
+ "val/CrossTalk.sources": null,
266
+ "CrossTalk.weights": null,
267
+ "train/CrossTalk.weights": null,
268
+ "val/CrossTalk.weights": null,
269
+ "CrossTalk.name": null,
270
+ "train/CrossTalk.name": null,
271
+ "val/CrossTalk.name": null,
272
+ "CrossTalk.prob": 1.0,
273
+ "train/CrossTalk.prob": 1.0,
274
+ "val/CrossTalk.prob": 1.0,
275
+ "CrossTalk.loudness_cutoff": -40,
276
+ "train/CrossTalk.loudness_cutoff": -40,
277
+ "val/CrossTalk.loudness_cutoff": -40,
278
+ "Equalizer.eq_amount": [
279
+ "const",
280
+ 1.0
281
+ ],
282
+ "train/Equalizer.eq_amount": [
283
+ "const",
284
+ 1.0
285
+ ],
286
+ "val/Equalizer.eq_amount": [
287
+ "const",
288
+ 1.0
289
+ ],
290
+ "Equalizer.n_bands": 6,
291
+ "train/Equalizer.n_bands": 6,
292
+ "val/Equalizer.n_bands": 6,
293
+ "Equalizer.name": null,
294
+ "train/Equalizer.name": null,
295
+ "val/Equalizer.name": null,
296
+ "Equalizer.prob": 1.0,
297
+ "train/Equalizer.prob": 1.0,
298
+ "val/Equalizer.prob": 1.0,
299
+ "FrequencyMask.f_center": [
300
+ "uniform",
301
+ 0.0,
302
+ 1.0
303
+ ],
304
+ "train/FrequencyMask.f_center": [
305
+ "uniform",
306
+ 0.0,
307
+ 1.0
308
+ ],
309
+ "val/FrequencyMask.f_center": [
310
+ "uniform",
311
+ 0.0,
312
+ 1.0
313
+ ],
314
+ "FrequencyMask.f_width": [
315
+ "const",
316
+ 0.1
317
+ ],
318
+ "train/FrequencyMask.f_width": [
319
+ "const",
320
+ 0.1
321
+ ],
322
+ "val/FrequencyMask.f_width": [
323
+ "const",
324
+ 0.1
325
+ ],
326
+ "FrequencyMask.name": null,
327
+ "train/FrequencyMask.name": null,
328
+ "val/FrequencyMask.name": null,
329
+ "FrequencyMask.prob": 1,
330
+ "train/FrequencyMask.prob": 1,
331
+ "val/FrequencyMask.prob": 1,
332
+ "FrequencyNoise.f_center": [
333
+ "uniform",
334
+ 0.0,
335
+ 1.0
336
+ ],
337
+ "train/FrequencyNoise.f_center": [
338
+ "uniform",
339
+ 0.0,
340
+ 1.0
341
+ ],
342
+ "val/FrequencyNoise.f_center": [
343
+ "uniform",
344
+ 0.0,
345
+ 1.0
346
+ ],
347
+ "FrequencyNoise.f_width": [
348
+ "const",
349
+ 0.1
350
+ ],
351
+ "train/FrequencyNoise.f_width": [
352
+ "const",
353
+ 0.1
354
+ ],
355
+ "val/FrequencyNoise.f_width": [
356
+ "const",
357
+ 0.1
358
+ ],
359
+ "FrequencyNoise.name": null,
360
+ "train/FrequencyNoise.name": null,
361
+ "val/FrequencyNoise.name": null,
362
+ "FrequencyNoise.prob": 1,
363
+ "train/FrequencyNoise.prob": 1,
364
+ "val/FrequencyNoise.prob": 1,
365
+ "GlobalVolumeNorm.db": [
366
+ "const",
367
+ -24
368
+ ],
369
+ "train/GlobalVolumeNorm.db": [
370
+ "const",
371
+ -24
372
+ ],
373
+ "val/GlobalVolumeNorm.db": [
374
+ "const",
375
+ -24
376
+ ],
377
+ "GlobalVolumeNorm.name": null,
378
+ "train/GlobalVolumeNorm.name": null,
379
+ "val/GlobalVolumeNorm.name": null,
380
+ "GlobalVolumeNorm.prob": 1.0,
381
+ "train/GlobalVolumeNorm.prob": 1.0,
382
+ "val/GlobalVolumeNorm.prob": 1.0,
383
+ "HighPass.cutoff": [
384
+ "choice",
385
+ [
386
+ 50,
387
+ 100,
388
+ 250,
389
+ 500,
390
+ 1000
391
+ ]
392
+ ],
393
+ "train/HighPass.cutoff": [
394
+ "choice",
395
+ [
396
+ 50,
397
+ 100,
398
+ 250,
399
+ 500,
400
+ 1000
401
+ ]
402
+ ],
403
+ "val/HighPass.cutoff": [
404
+ "choice",
405
+ [
406
+ 50,
407
+ 100,
408
+ 250,
409
+ 500,
410
+ 1000
411
+ ]
412
+ ],
413
+ "HighPass.zeros": 51,
414
+ "train/HighPass.zeros": 51,
415
+ "val/HighPass.zeros": 51,
416
+ "HighPass.name": null,
417
+ "train/HighPass.name": null,
418
+ "val/HighPass.name": null,
419
+ "HighPass.prob": 1,
420
+ "train/HighPass.prob": 1,
421
+ "val/HighPass.prob": 1,
422
+ "BaseTransform.keys": [],
423
+ "train/BaseTransform.keys": [],
424
+ "val/BaseTransform.keys": [],
425
+ "BaseTransform.name": null,
426
+ "train/BaseTransform.name": null,
427
+ "val/BaseTransform.name": null,
428
+ "BaseTransform.prob": 1.0,
429
+ "train/BaseTransform.prob": 1.0,
430
+ "val/BaseTransform.prob": 1.0,
431
+ "InvertPhase.name": null,
432
+ "train/InvertPhase.name": null,
433
+ "val/InvertPhase.name": null,
434
+ "InvertPhase.prob": 1,
435
+ "train/InvertPhase.prob": 1,
436
+ "val/InvertPhase.prob": 1,
437
+ "LowPass.cutoff": [
438
+ "choice",
439
+ [
440
+ 4000,
441
+ 8000,
442
+ 16000
443
+ ]
444
+ ],
445
+ "train/LowPass.cutoff": [
446
+ "choice",
447
+ [
448
+ 4000,
449
+ 8000,
450
+ 16000
451
+ ]
452
+ ],
453
+ "val/LowPass.cutoff": [
454
+ "choice",
455
+ [
456
+ 4000,
457
+ 8000,
458
+ 16000
459
+ ]
460
+ ],
461
+ "LowPass.zeros": 51,
462
+ "train/LowPass.zeros": 51,
463
+ "val/LowPass.zeros": 51,
464
+ "LowPass.name": null,
465
+ "train/LowPass.name": null,
466
+ "val/LowPass.name": null,
467
+ "LowPass.prob": 1,
468
+ "train/LowPass.prob": 1,
469
+ "val/LowPass.prob": 1,
470
+ "MaskLowMagnitudes.db_cutoff": [
471
+ "uniform",
472
+ -10,
473
+ 10
474
+ ],
475
+ "train/MaskLowMagnitudes.db_cutoff": [
476
+ "uniform",
477
+ -10,
478
+ 10
479
+ ],
480
+ "val/MaskLowMagnitudes.db_cutoff": [
481
+ "uniform",
482
+ -10,
483
+ 10
484
+ ],
485
+ "MaskLowMagnitudes.name": null,
486
+ "train/MaskLowMagnitudes.name": null,
487
+ "val/MaskLowMagnitudes.name": null,
488
+ "MaskLowMagnitudes.prob": 1,
489
+ "train/MaskLowMagnitudes.prob": 1,
490
+ "val/MaskLowMagnitudes.prob": 1,
491
+ "MuLawQuantization.channels": [
492
+ "choice",
493
+ [
494
+ 8,
495
+ 32,
496
+ 128,
497
+ 256,
498
+ 1024
499
+ ]
500
+ ],
501
+ "train/MuLawQuantization.channels": [
502
+ "choice",
503
+ [
504
+ 8,
505
+ 32,
506
+ 128,
507
+ 256,
508
+ 1024
509
+ ]
510
+ ],
511
+ "val/MuLawQuantization.channels": [
512
+ "choice",
513
+ [
514
+ 8,
515
+ 32,
516
+ 128,
517
+ 256,
518
+ 1024
519
+ ]
520
+ ],
521
+ "MuLawQuantization.name": null,
522
+ "train/MuLawQuantization.name": null,
523
+ "val/MuLawQuantization.name": null,
524
+ "MuLawQuantization.prob": 1.0,
525
+ "train/MuLawQuantization.prob": 1.0,
526
+ "val/MuLawQuantization.prob": 1.0,
527
+ "NoiseFloor.db": [
528
+ "const",
529
+ -50.0
530
+ ],
531
+ "train/NoiseFloor.db": [
532
+ "const",
533
+ -50.0
534
+ ],
535
+ "val/NoiseFloor.db": [
536
+ "const",
537
+ -50.0
538
+ ],
539
+ "NoiseFloor.name": null,
540
+ "train/NoiseFloor.name": null,
541
+ "val/NoiseFloor.name": null,
542
+ "NoiseFloor.prob": 1.0,
543
+ "train/NoiseFloor.prob": 1.0,
544
+ "val/NoiseFloor.prob": 1.0,
545
+ "Quantization.channels": [
546
+ "choice",
547
+ [
548
+ 8,
549
+ 32,
550
+ 128,
551
+ 256,
552
+ 1024
553
+ ]
554
+ ],
555
+ "train/Quantization.channels": [
556
+ "choice",
557
+ [
558
+ 8,
559
+ 32,
560
+ 128,
561
+ 256,
562
+ 1024
563
+ ]
564
+ ],
565
+ "val/Quantization.channels": [
566
+ "choice",
567
+ [
568
+ 8,
569
+ 32,
570
+ 128,
571
+ 256,
572
+ 1024
573
+ ]
574
+ ],
575
+ "Quantization.name": null,
576
+ "train/Quantization.name": null,
577
+ "val/Quantization.name": null,
578
+ "Quantization.prob": 1.0,
579
+ "train/Quantization.prob": 1.0,
580
+ "val/Quantization.prob": 1.0,
581
+ "Repeat.n_repeat": 1,
582
+ "train/Repeat.n_repeat": 1,
583
+ "val/Repeat.n_repeat": 1,
584
+ "Repeat.name": null,
585
+ "train/Repeat.name": null,
586
+ "val/Repeat.name": null,
587
+ "Repeat.prob": 1.0,
588
+ "train/Repeat.prob": 1.0,
589
+ "val/Repeat.prob": 1.0,
590
+ "RepeatUpTo.max_repeat": 5,
591
+ "train/RepeatUpTo.max_repeat": 5,
592
+ "val/RepeatUpTo.max_repeat": 5,
593
+ "RepeatUpTo.weights": null,
594
+ "train/RepeatUpTo.weights": null,
595
+ "val/RepeatUpTo.weights": null,
596
+ "RepeatUpTo.name": null,
597
+ "train/RepeatUpTo.name": null,
598
+ "val/RepeatUpTo.name": null,
599
+ "RepeatUpTo.prob": 1.0,
600
+ "train/RepeatUpTo.prob": 1.0,
601
+ "val/RepeatUpTo.prob": 1.0,
602
+ "RescaleAudio.val": 1.0,
603
+ "train/RescaleAudio.val": 1.0,
604
+ "val/RescaleAudio.val": 1.0,
605
+ "RescaleAudio.name": null,
606
+ "train/RescaleAudio.name": null,
607
+ "val/RescaleAudio.name": null,
608
+ "RescaleAudio.prob": 1,
609
+ "train/RescaleAudio.prob": 1,
610
+ "val/RescaleAudio.prob": 1,
611
+ "RoomImpulseResponse.drr": [
612
+ "uniform",
613
+ 0.0,
614
+ 30.0
615
+ ],
616
+ "train/RoomImpulseResponse.drr": [
617
+ "uniform",
618
+ 0.0,
619
+ 30.0
620
+ ],
621
+ "val/RoomImpulseResponse.drr": [
622
+ "uniform",
623
+ 0.0,
624
+ 30.0
625
+ ],
626
+ "RoomImpulseResponse.sources": null,
627
+ "train/RoomImpulseResponse.sources": null,
628
+ "val/RoomImpulseResponse.sources": null,
629
+ "RoomImpulseResponse.weights": null,
630
+ "train/RoomImpulseResponse.weights": null,
631
+ "val/RoomImpulseResponse.weights": null,
632
+ "RoomImpulseResponse.eq_amount": [
633
+ "const",
634
+ 1.0
635
+ ],
636
+ "train/RoomImpulseResponse.eq_amount": [
637
+ "const",
638
+ 1.0
639
+ ],
640
+ "val/RoomImpulseResponse.eq_amount": [
641
+ "const",
642
+ 1.0
643
+ ],
644
+ "RoomImpulseResponse.n_bands": 6,
645
+ "train/RoomImpulseResponse.n_bands": 6,
646
+ "val/RoomImpulseResponse.n_bands": 6,
647
+ "RoomImpulseResponse.name": null,
648
+ "train/RoomImpulseResponse.name": null,
649
+ "val/RoomImpulseResponse.name": null,
650
+ "RoomImpulseResponse.prob": 1.0,
651
+ "train/RoomImpulseResponse.prob": 1.0,
652
+ "val/RoomImpulseResponse.prob": 1.0,
653
+ "RoomImpulseResponse.use_original_phase": false,
654
+ "train/RoomImpulseResponse.use_original_phase": false,
655
+ "val/RoomImpulseResponse.use_original_phase": false,
656
+ "RoomImpulseResponse.offset": 0.0,
657
+ "train/RoomImpulseResponse.offset": 0.0,
658
+ "val/RoomImpulseResponse.offset": 0.0,
659
+ "RoomImpulseResponse.duration": 1.0,
660
+ "train/RoomImpulseResponse.duration": 1.0,
661
+ "val/RoomImpulseResponse.duration": 1.0,
662
+ "ShiftPhase.shift": [
663
+ "uniform",
664
+ -3.141592653589793,
665
+ 3.141592653589793
666
+ ],
667
+ "train/ShiftPhase.shift": [
668
+ "uniform",
669
+ -3.141592653589793,
670
+ 3.141592653589793
671
+ ],
672
+ "val/ShiftPhase.shift": [
673
+ "uniform",
674
+ -3.141592653589793,
675
+ 3.141592653589793
676
+ ],
677
+ "ShiftPhase.name": null,
678
+ "train/ShiftPhase.name": null,
679
+ "val/ShiftPhase.name": null,
680
+ "ShiftPhase.prob": 1,
681
+ "train/ShiftPhase.prob": 1,
682
+ "val/ShiftPhase.prob": 1,
683
+ "Silence.name": null,
684
+ "train/Silence.name": null,
685
+ "val/Silence.name": null,
686
+ "Silence.prob": 0.1,
687
+ "train/Silence.prob": 0.1,
688
+ "val/Silence.prob": 0.1,
689
+ "Smoothing.window_type": [
690
+ "const",
691
+ "average"
692
+ ],
693
+ "train/Smoothing.window_type": [
694
+ "const",
695
+ "average"
696
+ ],
697
+ "val/Smoothing.window_type": [
698
+ "const",
699
+ "average"
700
+ ],
701
+ "Smoothing.window_length": [
702
+ "choice",
703
+ [
704
+ 8,
705
+ 16,
706
+ 32,
707
+ 64,
708
+ 128,
709
+ 256,
710
+ 512
711
+ ]
712
+ ],
713
+ "train/Smoothing.window_length": [
714
+ "choice",
715
+ [
716
+ 8,
717
+ 16,
718
+ 32,
719
+ 64,
720
+ 128,
721
+ 256,
722
+ 512
723
+ ]
724
+ ],
725
+ "val/Smoothing.window_length": [
726
+ "choice",
727
+ [
728
+ 8,
729
+ 16,
730
+ 32,
731
+ 64,
732
+ 128,
733
+ 256,
734
+ 512
735
+ ]
736
+ ],
737
+ "Smoothing.name": null,
738
+ "train/Smoothing.name": null,
739
+ "val/Smoothing.name": null,
740
+ "Smoothing.prob": 1,
741
+ "train/Smoothing.prob": 1,
742
+ "val/Smoothing.prob": 1,
743
+ "SpectralDenoising.eq_amount": [
744
+ "const",
745
+ 1.0
746
+ ],
747
+ "train/SpectralDenoising.eq_amount": [
748
+ "const",
749
+ 1.0
750
+ ],
751
+ "val/SpectralDenoising.eq_amount": [
752
+ "const",
753
+ 1.0
754
+ ],
755
+ "SpectralDenoising.denoise_amount": [
756
+ "uniform",
757
+ 0.8,
758
+ 1.0
759
+ ],
760
+ "train/SpectralDenoising.denoise_amount": [
761
+ "uniform",
762
+ 0.8,
763
+ 1.0
764
+ ],
765
+ "val/SpectralDenoising.denoise_amount": [
766
+ "uniform",
767
+ 0.8,
768
+ 1.0
769
+ ],
770
+ "SpectralDenoising.nz_volume": -40,
771
+ "train/SpectralDenoising.nz_volume": -40,
772
+ "val/SpectralDenoising.nz_volume": -40,
773
+ "SpectralDenoising.n_bands": 6,
774
+ "train/SpectralDenoising.n_bands": 6,
775
+ "val/SpectralDenoising.n_bands": 6,
776
+ "SpectralDenoising.n_freq": 3,
777
+ "train/SpectralDenoising.n_freq": 3,
778
+ "val/SpectralDenoising.n_freq": 3,
779
+ "SpectralDenoising.n_time": 5,
780
+ "train/SpectralDenoising.n_time": 5,
781
+ "val/SpectralDenoising.n_time": 5,
782
+ "SpectralDenoising.name": null,
783
+ "train/SpectralDenoising.name": null,
784
+ "val/SpectralDenoising.name": null,
785
+ "SpectralDenoising.prob": 1,
786
+ "train/SpectralDenoising.prob": 1,
787
+ "val/SpectralDenoising.prob": 1,
788
+ "TimeMask.t_center": [
789
+ "uniform",
790
+ 0.0,
791
+ 1.0
792
+ ],
793
+ "train/TimeMask.t_center": [
794
+ "uniform",
795
+ 0.0,
796
+ 1.0
797
+ ],
798
+ "val/TimeMask.t_center": [
799
+ "uniform",
800
+ 0.0,
801
+ 1.0
802
+ ],
803
+ "TimeMask.t_width": [
804
+ "const",
805
+ 0.025
806
+ ],
807
+ "train/TimeMask.t_width": [
808
+ "const",
809
+ 0.025
810
+ ],
811
+ "val/TimeMask.t_width": [
812
+ "const",
813
+ 0.025
814
+ ],
815
+ "TimeMask.name": null,
816
+ "train/TimeMask.name": null,
817
+ "val/TimeMask.name": null,
818
+ "TimeMask.prob": 1,
819
+ "train/TimeMask.prob": 1,
820
+ "val/TimeMask.prob": 1,
821
+ "TimeNoise.t_center": [
822
+ "uniform",
823
+ 0.0,
824
+ 1.0
825
+ ],
826
+ "train/TimeNoise.t_center": [
827
+ "uniform",
828
+ 0.0,
829
+ 1.0
830
+ ],
831
+ "val/TimeNoise.t_center": [
832
+ "uniform",
833
+ 0.0,
834
+ 1.0
835
+ ],
836
+ "TimeNoise.t_width": [
837
+ "const",
838
+ 0.025
839
+ ],
840
+ "train/TimeNoise.t_width": [
841
+ "const",
842
+ 0.025
843
+ ],
844
+ "val/TimeNoise.t_width": [
845
+ "const",
846
+ 0.025
847
+ ],
848
+ "TimeNoise.name": null,
849
+ "train/TimeNoise.name": null,
850
+ "val/TimeNoise.name": null,
851
+ "TimeNoise.prob": 1,
852
+ "train/TimeNoise.prob": 1,
853
+ "val/TimeNoise.prob": 1,
854
+ "VolumeChange.db": [
855
+ "uniform",
856
+ -12.0,
857
+ 0.0
858
+ ],
859
+ "train/VolumeChange.db": [
860
+ "uniform",
861
+ -12.0,
862
+ 0.0
863
+ ],
864
+ "val/VolumeChange.db": [
865
+ "uniform",
866
+ -12.0,
867
+ 0.0
868
+ ],
869
+ "VolumeChange.name": null,
870
+ "train/VolumeChange.name": null,
871
+ "val/VolumeChange.name": null,
872
+ "VolumeChange.prob": 1.0,
873
+ "train/VolumeChange.prob": 1.0,
874
+ "val/VolumeChange.prob": 1.0,
875
+ "VolumeNorm.db": [
876
+ "const",
877
+ -16
878
+ ],
879
+ "train/VolumeNorm.db": [
880
+ "const",
881
+ -16
882
+ ],
883
+ "val/VolumeNorm.db": [
884
+ "const",
885
+ -16
886
+ ],
887
+ "VolumeNorm.name": null,
888
+ "train/VolumeNorm.name": null,
889
+ "val/VolumeNorm.name": null,
890
+ "VolumeNorm.prob": 1.0,
891
+ "train/VolumeNorm.prob": 1.0,
892
+ "val/VolumeNorm.prob": 1.0,
893
+ "L1Loss.attribute": "audio_data",
894
+ "L1Loss.weight": 1.0,
895
+ "MelSpectrogramLoss.n_mels": [
896
+ 5,
897
+ 10,
898
+ 20,
899
+ 40,
900
+ 80,
901
+ 160,
902
+ 320
903
+ ],
904
+ "MelSpectrogramLoss.window_lengths": [
905
+ 32,
906
+ 64,
907
+ 128,
908
+ 256,
909
+ 512,
910
+ 1024,
911
+ 2048
912
+ ],
913
+ "MelSpectrogramLoss.clamp_eps": 1e-05,
914
+ "MelSpectrogramLoss.mag_weight": 0.0,
915
+ "MelSpectrogramLoss.log_weight": 1.0,
916
+ "MelSpectrogramLoss.pow": 1.0,
917
+ "MelSpectrogramLoss.weight": 1.0,
918
+ "MelSpectrogramLoss.match_stride": false,
919
+ "MelSpectrogramLoss.mel_fmin": [
920
+ 0,
921
+ 0,
922
+ 0,
923
+ 0,
924
+ 0,
925
+ 0,
926
+ 0
927
+ ],
928
+ "MelSpectrogramLoss.mel_fmax": [
929
+ null,
930
+ null,
931
+ null,
932
+ null,
933
+ null,
934
+ null,
935
+ null
936
+ ],
937
+ "MelSpectrogramLoss.window_type": null,
938
+ "MultiScaleSTFTLoss.window_lengths": [
939
+ 2048,
940
+ 512
941
+ ],
942
+ "MultiScaleSTFTLoss.clamp_eps": 1e-05,
943
+ "MultiScaleSTFTLoss.mag_weight": 1.0,
944
+ "MultiScaleSTFTLoss.log_weight": 1.0,
945
+ "MultiScaleSTFTLoss.pow": 2.0,
946
+ "MultiScaleSTFTLoss.weight": 1.0,
947
+ "MultiScaleSTFTLoss.match_stride": false,
948
+ "MultiScaleSTFTLoss.window_type": null,
949
+ "SISDRLoss.scaling": true,
950
+ "SISDRLoss.reduction": "mean",
951
+ "SISDRLoss.zero_mean": true,
952
+ "SISDRLoss.clip_min": null,
953
+ "SISDRLoss.weight": 1.0,
954
+ "build_transform.augment_prob": 0.0,
955
+ "train/build_transform.augment_prob": 0.0,
956
+ "val/build_transform.augment_prob": 1.0,
957
+ "build_transform.preprocess": [
958
+ "Identity"
959
+ ],
960
+ "train/build_transform.preprocess": [
961
+ "Identity"
962
+ ],
963
+ "val/build_transform.preprocess": [
964
+ "Identity"
965
+ ],
966
+ "build_transform.augment": [
967
+ "Identity"
968
+ ],
969
+ "train/build_transform.augment": [
970
+ "Identity"
971
+ ],
972
+ "val/build_transform.augment": [
973
+ "Identity"
974
+ ],
975
+ "build_transform.postprocess": [
976
+ "VolumeNorm",
977
+ "RescaleAudio",
978
+ "ShiftPhase"
979
+ ],
980
+ "train/build_transform.postprocess": [
981
+ "VolumeNorm",
982
+ "RescaleAudio",
983
+ "ShiftPhase"
984
+ ],
985
+ "val/build_transform.postprocess": [
986
+ "VolumeNorm",
987
+ "RescaleAudio",
988
+ "ShiftPhase"
989
+ ],
990
+ "build_dataset.folders": null,
991
+ "train/build_dataset.folders": {
992
+ "speech": [
993
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/public/public_datas/speech/LibriTTS/train-clean-100",
994
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/public/public_datas/speech/LibriTTS/train-clean-360",
995
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/public/public_datas/speech/LibriTTS/train-other-500",
996
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/public/public_datas/speech/Libriheavy/medium",
997
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/public/public_datas/speech/Libriheavy/small"
998
+ ]
999
+ },
1000
+ "val/build_dataset.folders": {
1001
+ "speech": [
1002
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/public/public_datas/speech/LibriTTS/dev-clean"
1003
+ ]
1004
+ },
1005
+ "test/build_dataset.folders": {
1006
+ "speech": [
1007
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/public/public_datas/speech/LibriTTS/test-clean"
1008
+ ]
1009
+ },
1010
+ "resume": false,
1011
+ "tag": "latest",
1012
+ "load_weights": false,
1013
+ "seed": 0,
1014
+ "save_path": "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/chenxie-25019/svae/ckpts/svae/16khz_40hz_5544_kl1e-2_vae64_7khr_bigvgan_base_ll_sm_attn_proj1_align_vae_mlp",
1015
+ "num_iters": 600001,
1016
+ "save_iters": [
1017
+ 100000,
1018
+ 200000,
1019
+ 300000,
1020
+ 400000,
1021
+ 500000,
1022
+ 600000
1023
+ ],
1024
+ "sample_freq": 10000,
1025
+ "valid_freq": 1000,
1026
+ "batch_size": 40,
1027
+ "val_batch_size": 8,
1028
+ "num_workers": 48,
1029
+ "val_idx": [
1030
+ 0,
1031
+ 1,
1032
+ 2,
1033
+ 3,
1034
+ 4,
1035
+ 5,
1036
+ 6,
1037
+ 7
1038
+ ],
1039
+ "lambdas": {
1040
+ "mel/loss": 15.0,
1041
+ "adv/feat_loss": 2.0,
1042
+ "adv/gen_loss": 1.0,
1043
+ "vae/kl_loss": 0.01,
1044
+ "vae/proj_loss": 1.0
1045
+ },
1046
+ "use_kl_warmup": false,
1047
+ "kl_warmup_ratio": 0.4,
1048
+ "kl_start_weight": 1e-05,
1049
+ "args.unknown": [],
1050
+ "DAC.vae_dim": 64,
1051
+ "AdamW.lr": 0.0001,
1052
+ "device": "cuda",
1053
+ "train/AudioDataset.guidance_path": [
1054
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/chenxie-25019/svae/scripts/extract_ssl_feature/hubert_large_66k_last_layer/LibriTTS/train-clean-100",
1055
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/chenxie-25019/svae/scripts/extract_ssl_feature/hubert_large_66k_last_layer/LibriTTS/train-clean-360",
1056
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/chenxie-25019/svae/scripts/extract_ssl_feature/hubert_large_66k_last_layer/LibriTTS/train-other-500",
1057
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/chenxie-25019/svae/scripts/extract_ssl_feature/hubert_large_66k_last_layer/LibriLight/medium",
1058
+ "/inspire/hdd/ws-f4d69b29-e0a5-44e6-bd92-acf4de9990f0/public-project/chenxie-25019/svae/scripts/extract_ssl_feature/hubert_large_66k_last_layer/LibriLight/small"
1059
+ ],
1060
+ "test/AudioDataset.duration": 10.0,
1061
+ "test/build_transform.augment_prob": 1.0,
1062
+ "test/AudioDataset.n_examples": 1000,
1063
+ "args.debug": 0,
1064
+ "args.load": "conf/svae_sii/16khz_40hz_5544_kl1e-2_vae64_7khr_bigvgan_base_ll_sm_attn_proj_align_vae.yml",
1065
+ "args.save": null
1066
+ }
svae/dac/ema_state_dict.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25a5fedf12f6bfe5de02d54f30813c2f1c5d6487322a69dea228c61c17624bd4
3
+ size 321925784
svae/dac/weights.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:de58436179fb43857333bc98a0a56cdf77e65d38b82194730d4cd7a53b4075e9
3
+ size 321864242
svae/metainfo.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"DAC": {"encoder_dim": 64, "encoder_rates": [4, 4, 5, 5], "latent_dim": null, "decoder_dim": 1024, "decoder_rates": [5, 5, 4, 4], "sample_rate": 16000, "distill": true, "distill_hidden_dim": 1024, "decoder_type": "bigvgan", "pre_vae_block": false, "attn_proj": true, "post_vae_block": false, "bigvgan_conf": "conf/bigvgan_conf/bigvgan_v2_16khz_40hz_base_vae64.json", "align_space": "vae", "sampling_ratios": [0, 1], "vae_dim": 64}, "Discriminator": {"rates": [], "periods": [2, 3, 5, 7, 11], "fft_sizes": [2048, 1024, 512], "sample_rate": 16000, "bands": [[0.0, 0.1], [0.1, 0.25], [0.25, 0.5], [0.5, 0.75], [0.75, 1.0]], "use_msstft_replace_mrd": false}}