Create mini_config.yaml
Browse files- mini_config.yaml +105 -0
mini_config.yaml
ADDED
|
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
log_dir: "Models/Output"
|
| 2 |
+
save_freq: 5
|
| 3 |
+
log_interval: 10
|
| 4 |
+
device: "cuda"
|
| 5 |
+
epochs: 50
|
| 6 |
+
batch_size: 8
|
| 7 |
+
max_len: 400
|
| 8 |
+
pretrained_model: ""
|
| 9 |
+
second_stage_load_pretrained: true
|
| 10 |
+
load_only_params: true
|
| 11 |
+
|
| 12 |
+
external_models:
|
| 13 |
+
asr:
|
| 14 |
+
input_dim: 80
|
| 15 |
+
hidden_dim: 256
|
| 16 |
+
n_token: 178
|
| 17 |
+
plbert:
|
| 18 |
+
vocab_size: 178
|
| 19 |
+
hidden_size: 768
|
| 20 |
+
num_attention_heads: 12
|
| 21 |
+
intermediate_size: 2048
|
| 22 |
+
dropout: 0.1
|
| 23 |
+
|
| 24 |
+
data_params:
|
| 25 |
+
train_data: "shethjenil/audiodata"
|
| 26 |
+
root_path: ""
|
| 27 |
+
min_length: 50
|
| 28 |
+
|
| 29 |
+
preprocess_params:
|
| 30 |
+
sr: 24000
|
| 31 |
+
n_fft: 2048
|
| 32 |
+
win_length: 1200
|
| 33 |
+
hop_length: 300
|
| 34 |
+
|
| 35 |
+
model_params:
|
| 36 |
+
multispeaker: true
|
| 37 |
+
dim_in: 64
|
| 38 |
+
hidden_dim: 128
|
| 39 |
+
max_conv_dim: 512
|
| 40 |
+
n_layer: 2
|
| 41 |
+
n_mels: 80
|
| 42 |
+
n_token: 178
|
| 43 |
+
max_dur: 50
|
| 44 |
+
style_dim: 128
|
| 45 |
+
dropout: 0.2
|
| 46 |
+
decoder:
|
| 47 |
+
type: "istftnet"
|
| 48 |
+
hidden_dim: 256
|
| 49 |
+
decoder_out_dim: 256
|
| 50 |
+
asr_res_in: 128
|
| 51 |
+
resblock_kernel_sizes: [3, 3]
|
| 52 |
+
upsample_rates: [10, 6]
|
| 53 |
+
upsample_initial_channel: 256
|
| 54 |
+
resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5], [1, 3, 5]]
|
| 55 |
+
upsample_kernel_sizes: [20, 12]
|
| 56 |
+
gen_istft_n_fft: 20
|
| 57 |
+
gen_istft_hop_size: 5
|
| 58 |
+
disable_complex: true
|
| 59 |
+
slm:
|
| 60 |
+
model: "microsoft/wavlm-base-plus"
|
| 61 |
+
sr: 16000
|
| 62 |
+
hidden: 768
|
| 63 |
+
nlayers: 13
|
| 64 |
+
initial_channel: 64
|
| 65 |
+
diffusion:
|
| 66 |
+
embedding_mask_proba: 0.1
|
| 67 |
+
transformer:
|
| 68 |
+
num_layers: 3
|
| 69 |
+
num_heads: 8
|
| 70 |
+
head_features: 64
|
| 71 |
+
multiplier: 2
|
| 72 |
+
dist:
|
| 73 |
+
sigma_data: 0.2
|
| 74 |
+
estimate_sigma_data: true
|
| 75 |
+
mean: -3.0
|
| 76 |
+
std: 1.0
|
| 77 |
+
|
| 78 |
+
loss_params:
|
| 79 |
+
lambda_mel: 5.0
|
| 80 |
+
lambda_gen: 1.0
|
| 81 |
+
lambda_slm: 1.0
|
| 82 |
+
lambda_mono: 1.0
|
| 83 |
+
lambda_s2s: 1.0
|
| 84 |
+
lambda_f0: 1.0
|
| 85 |
+
lambda_norm: 1.0
|
| 86 |
+
lambda_dur: 1.0
|
| 87 |
+
lambda_ce: 20.0
|
| 88 |
+
lambda_sty: 1.0
|
| 89 |
+
lambda_diff: 1.0
|
| 90 |
+
diff_epoch: 10
|
| 91 |
+
joint_epoch: 30
|
| 92 |
+
|
| 93 |
+
optimizer_params:
|
| 94 |
+
lr: 0.0001
|
| 95 |
+
bert_lr: 0.00001
|
| 96 |
+
ft_lr: 0.0001
|
| 97 |
+
|
| 98 |
+
slmadv_params:
|
| 99 |
+
min_len: 400
|
| 100 |
+
max_len: 500
|
| 101 |
+
batch_percentage: 0.5
|
| 102 |
+
iter: 10
|
| 103 |
+
thresh: 5.0
|
| 104 |
+
scale: 0.01
|
| 105 |
+
sig: 1.5
|