update

egs/egs_bases/tts/vocoder/hifigan.yaml

@@ -0,0 +1,38 @@
+base_config: ./base.yaml
+task_cls: tasks.vocoder.hifigan.HifiGanTask
+resblock: "1"
+adam_b1: 0.8
+adam_b2: 0.99
+upsample_rates: [ 8,8,2,2 ]
+upsample_kernel_sizes: [ 16,16,4,4 ]
+upsample_initial_channel: 512
+resblock_kernel_sizes: [ 3,7,11 ]
+resblock_dilation_sizes: [ [ 1,3,5 ], [ 1,3,5 ], [ 1,3,5 ] ]
+
+use_pitch_embed: false
+use_fm_loss: false
+use_spec_disc: false
+use_cdisc: false
+use_ms_stft: false
+use_cond_disc: false
+
+lambda_mel: 5.0
+lambda_mel_adv: 1.0
+lambda_cdisc: 4.0
+lambda_adv: 1.0
+
+max_samples: 8192
+max_sentences: 24
+
+aux_context_window: 0 # Context window size for auxiliary feature.
+generator_optimizer_params:
+  lr: 0.0002            # Generator's learning rate.
+generator_scheduler_params:
+  step_size: 600
+  gamma: 0.999
+discriminator_optimizer_params:
+  lr: 0.0002            # Discriminator's learning rate.
+discriminator_scheduler_params:
+  step_size: 600
+  gamma: 0.999
+max_updates: 3000000