| allow_cache: true | |
| batch_size: 32 | |
| end_ratio_value: 0.0 | |
| eval_interval_steps: 500 | |
| format: npy | |
| gradient_accumulation_steps: 1 | |
| hop_size: 256 | |
| is_shuffle: true | |
| log_interval_steps: 200 | |
| mel_length_threshold: 32 | |
| model_type: tacotron2 | |
| num_save_intermediate_results: 1 | |
| optimizer_params: | |
| decay_steps: 150000 | |
| end_learning_rate: 1.0e-05 | |
| initial_learning_rate: 0.001 | |
| warmup_proportion: 0.02 | |
| weight_decay: 0.001 | |
| remove_short_samples: true | |
| save_interval_steps: 2000 | |
| schedule_decay_steps: 50000 | |
| start_ratio_value: 0.5 | |
| start_schedule_teacher_forcing: 200001 | |
| tacotron2_params: | |
| attention_dim: 128 | |
| attention_filters: 32 | |
| attention_kernel: 31 | |
| attention_type: lsa | |
| dataset: kss | |
| decoder_lstm_units: 1024 | |
| embedding_dropout_prob: 0.1 | |
| embedding_hidden_size: 512 | |
| encoder_conv_activation: relu | |
| encoder_conv_dropout_rate: 0.5 | |
| encoder_conv_filters: 512 | |
| encoder_conv_kernel_sizes: 5 | |
| encoder_lstm_units: 256 | |
| initializer_range: 0.02 | |
| n_conv_encoder: 5 | |
| n_conv_postnet: 5 | |
| n_lstm_decoder: 1 | |
| n_mels: 80 | |
| n_prenet_layers: 2 | |
| n_speakers: 1 | |
| postnet_conv_filters: 512 | |
| postnet_conv_kernel_sizes: 5 | |
| postnet_dropout_rate: 0.1 | |
| prenet_activation: relu | |
| prenet_dropout_rate: 0.5 | |
| prenet_units: 256 | |
| reduction_factor: 1 | |
| train_max_steps: 200000 | |
| use_fixed_shapes: true | |
| var_train_expr: null | |