Upload config.yml
Browse files- config.yml +43 -0
config.yml
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
allow_cache: true
|
| 2 |
+
batch_size: 1
|
| 3 |
+
config: ./ttsexamples/tacotron2/conf/tacotron2.v1.yaml
|
| 4 |
+
dev_dir: /content/dump_ljspeech/valid
|
| 5 |
+
end_ratio_value: 0.0
|
| 6 |
+
eval_interval_steps: 50
|
| 7 |
+
format: npy
|
| 8 |
+
gradient_accumulation_steps: 1
|
| 9 |
+
hop_size: 256
|
| 10 |
+
is_shuffle: true
|
| 11 |
+
log_interval_steps: 20
|
| 12 |
+
max_char_length: 128
|
| 13 |
+
max_mel_length: 257
|
| 14 |
+
mel_length_threshold: 32
|
| 15 |
+
mixed_precision: false
|
| 16 |
+
model_type: tacotron2
|
| 17 |
+
num_save_intermediate_results: 1
|
| 18 |
+
optimizer_params: {decay_steps: 150000, end_learning_rate: 1.0e-05, initial_learning_rate: 0.001,
|
| 19 |
+
warmup_proportion: 0.02, weight_decay: 0.001}
|
| 20 |
+
outdir: ./ttsexamples/tacotron2/exp/train.tacotron2.v1/
|
| 21 |
+
pretrained: ''
|
| 22 |
+
remove_short_samples: true
|
| 23 |
+
resume: ''
|
| 24 |
+
save_interval_steps: 20
|
| 25 |
+
schedule_decay_steps: 50000
|
| 26 |
+
start_ratio_value: 0.5
|
| 27 |
+
start_schedule_teacher_forcing: 200001
|
| 28 |
+
tacotron2_params: {attention_dim: 128, attention_filters: 32, attention_kernel: 31,
|
| 29 |
+
attention_type: lsa, dataset: ljspeech, decoder_lstm_units: 1024, embedding_dropout_prob: 0.1,
|
| 30 |
+
embedding_hidden_size: 512, encoder_conv_activation: relu, encoder_conv_dropout_rate: 0.5,
|
| 31 |
+
encoder_conv_filters: 512, encoder_conv_kernel_sizes: 5, encoder_lstm_units: 256,
|
| 32 |
+
initializer_range: 0.02, n_conv_encoder: 5, n_conv_postnet: 5, n_lstm_decoder: 1,
|
| 33 |
+
n_mels: 80, n_prenet_layers: 2, n_speakers: 1, postnet_conv_filters: 512, postnet_conv_kernel_sizes: 5,
|
| 34 |
+
postnet_dropout_rate: 0.1, prenet_activation: relu, prenet_dropout_rate: 0.5, prenet_units: 256,
|
| 35 |
+
reduction_factor: 1}
|
| 36 |
+
train_dir: /content/dump_ljspeech/train
|
| 37 |
+
train_max_steps: 200000
|
| 38 |
+
use_fal: false
|
| 39 |
+
use_fixed_shapes: true
|
| 40 |
+
use_norm: true
|
| 41 |
+
var_train_expr: null
|
| 42 |
+
verbose: 1
|
| 43 |
+
version: '0.0'
|