text_to_speech/egs/egs_bases/tts/fs_adv.yaml

base_config: ./fs.yaml
task_cls: tasks.tts.fs_adv.FastSpeechAdvTask

lambda_mel_adv: 0.05

disc_win_num: 3
mel_disc_hidden_size: 128
disc_norm: in
disc_reduction: stack
disc_interval: 1
disc_lr: 0.0001
disc_start_steps: 0
discriminator_scheduler_params:
  gamma: 0.5
  step_size: 40000
discriminator_optimizer_params:
  eps: 1.0e-06
  weight_decay: 0.0

scheduler: rsqrt
lr: 0.4 # edit from 1.

# copied from graph-porta-speech repo
predictor_grad: 0.02

max_tokens: 30000
num_sanity_val_steps: 0 # steps of validation at the beginning
ds_workers: 1

use_pitch_embed: false # Attention! no pitch predictor!
pitch_type: cwt # frame|ph|cwt