Add config/everyvoice-text-to-spec.yaml for Turkish
Browse files
config/everyvoice-text-to-spec.yaml
ADDED
|
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
VERSION: '1.0'
|
| 2 |
+
contact: {contact_email: david.guzman@mail.mcgill.ca, contact_name: David Guzman}
|
| 3 |
+
model:
|
| 4 |
+
decoder: {conv_kernel_size: 9, dropout: 0.2, feedforward_dim: 1024, heads: 2, input_dim: 256,
|
| 5 |
+
layers: 4}
|
| 6 |
+
encoder: {conv_kernel_size: 9, dropout: 0.2, feedforward_dim: 1024, heads: 2, input_dim: 256,
|
| 7 |
+
layers: 4}
|
| 8 |
+
learn_alignment: true
|
| 9 |
+
max_length: 1000
|
| 10 |
+
mel_loss: mse
|
| 11 |
+
multilingual: false
|
| 12 |
+
multispeaker: false
|
| 13 |
+
target_text_representation_level: characters
|
| 14 |
+
use_global_style_token_module: false
|
| 15 |
+
use_postnet: true
|
| 16 |
+
variance_predictors:
|
| 17 |
+
duration: {depthwise: true, dropout: 0.5, input_dim: 256, kernel_size: 3, loss: mse,
|
| 18 |
+
n_bins: 256, n_layers: 5}
|
| 19 |
+
energy: {depthwise: true, dropout: 0.5, input_dim: 256, kernel_size: 3, level: phone,
|
| 20 |
+
loss: mse, n_bins: 256, n_layers: 5}
|
| 21 |
+
pitch: {depthwise: true, dropout: 0.5, input_dim: 256, kernel_size: 3, level: phone,
|
| 22 |
+
loss: mse, n_bins: 256, n_layers: 5}
|
| 23 |
+
path_to_model_config_file: null
|
| 24 |
+
path_to_preprocessing_config_file: everyvoice-shared-data.yaml
|
| 25 |
+
path_to_text_config_file: everyvoice-shared-text.yaml
|
| 26 |
+
path_to_training_config_file: null
|
| 27 |
+
training:
|
| 28 |
+
attn_bin_loss_warmup_epochs: 100
|
| 29 |
+
attn_bin_loss_weight: 0.1
|
| 30 |
+
attn_ctc_loss_weight: 0.1
|
| 31 |
+
batch_size: 16
|
| 32 |
+
check_val_every_n_epoch: null
|
| 33 |
+
ckpt_epochs: 1
|
| 34 |
+
ckpt_steps: null
|
| 35 |
+
duration_loss_weight: 0.1
|
| 36 |
+
energy_loss_weight: 0.1
|
| 37 |
+
filelist_loader: everyvoice.utils.generic_dict_loader
|
| 38 |
+
finetune_checkpoint: null
|
| 39 |
+
logger: {name: FeaturePredictionExperiment, save_dir: ../logs_and_checkpoints, sub_dir_callable: everyvoice.utils.get_current_time,
|
| 40 |
+
version: base}
|
| 41 |
+
max_epochs: 1000
|
| 42 |
+
max_steps: 100000
|
| 43 |
+
mel_loss_weight: 1.0
|
| 44 |
+
optimizer:
|
| 45 |
+
betas: [0.9, 0.999]
|
| 46 |
+
eps: 1.0e-08
|
| 47 |
+
learning_rate: 0.001
|
| 48 |
+
name: noam
|
| 49 |
+
warmup_steps: 1000
|
| 50 |
+
weight_decay: 1.0e-06
|
| 51 |
+
pitch_loss_weight: 0.1
|
| 52 |
+
postnet_loss_weight: 1.0
|
| 53 |
+
save_top_k_ckpts: 5
|
| 54 |
+
train_data_workers: 4
|
| 55 |
+
training_filelist: ../preprocessed/training_filelist.psv
|
| 56 |
+
use_weighted_sampler: false
|
| 57 |
+
val_check_interval: 500
|
| 58 |
+
val_data_workers: 0
|
| 59 |
+
validation_filelist: ../preprocessed/validation_filelist.psv
|
| 60 |
+
vocoder_path: null
|