| allow_cache: true | |
| batch_max_steps: 8192 | |
| batch_max_steps_valid: 81920 | |
| batch_size: 1 | |
| config: ./ttsexamples/melgan_stft/conf/melgan_stft.v1.yaml | |
| dev_dir: /content/dump_ljspeech/valid | |
| discriminator_mixed_precision: false | |
| discriminator_optimizer_params: | |
| lr_fn: PiecewiseConstantDecay | |
| lr_params: | |
| boundaries: [0] | |
| values: [0.0001, 0.0001] | |
| discriminator_train_start_steps: 100000 | |
| eval_interval_steps: 5000 | |
| format: npy | |
| generator_mixed_precision: true | |
| generator_optimizer_params: | |
| lr_fn: PiecewiseConstantDecay | |
| lr_params: | |
| boundaries: [100000] | |
| values: [0.0005, 0.0001] | |
| gradient_accumulation_steps: 1 | |
| hop_size: 256 | |
| is_shuffle: true | |
| lambda_adv: 4.0 | |
| lambda_feat_match: 10.0 | |
| log_interval_steps: 200 | |
| melgan_discriminator_params: | |
| downsample_pooling: AveragePooling1D | |
| downsample_pooling_params: {pool_size: 4, strides: 2} | |
| downsample_scales: [4, 4, 4, 4] | |
| filters: 16 | |
| is_weight_norm: false | |
| kernel_sizes: [5, 3] | |
| max_downsample_filters: 1024 | |
| nonlinear_activation: LeakyReLU | |
| nonlinear_activation_params: {alpha: 0.2} | |
| out_channels: 1 | |
| scales: 3 | |
| melgan_generator_params: | |
| filters: 512 | |
| is_weight_norm: false | |
| kernel_size: 7 | |
| out_channels: 1 | |
| stack_kernel_size: 3 | |
| stacks: 3 | |
| upsample_scales: [8, 8, 2, 2] | |
| model_type: melgan_generator | |
| num_save_intermediate_results: 1 | |
| outdir: ./ttsexamples/melgan_stft/exp/train.melgan_stft.v1/ | |
| pretrained: '' | |
| remove_short_samples: true | |
| resume: '' | |
| sampling_rate: 22050 | |
| save_interval_steps: 200 | |
| stft_loss_params: | |
| fft_lengths: [1024, 2048, 512] | |
| frame_lengths: [600, 1200, 240] | |
| frame_steps: [120, 240, 50] | |
| train_dir: /content/dump_ljspeech/train | |
| train_max_steps: 4000000 | |
| use_norm: true | |
| verbose: 1 | |
| version: '0.0' | |