speech_sambert-hifigan_tts_zh-cn_16k/voices/zhibei_emo/voc/config.yaml

model_type: hifigan
Model:
###########################################################
#         GENERATOR NETWORK ARCHITECTURE SETTING          #
###########################################################
  Generator:
    params:
      in_channels: 80                       
      out_channels: 1                      
      channels: 256                       
      kernel_size: 7                     
      upsample_scales: [10, 5, 2, 2]        
      upsample_kernal_sizes: [20, 11, 4, 4] 
      resblock_kernel_sizes: [3, 7, 11]     
      resblock_dilations:                  
            - [1, 3, 5, 7]
            - [1, 3, 5, 7]
            - [1, 3, 5, 7]
      bias: true                           
      causal: true                           
      nonlinear_activation: "LeakyReLU"    
      nonlinear_activation_params:         
        negative_slope: 0.1
      use_weight_norm: true               
    optimizer:
      type: Adam
      params:
        lr: 2.0e-4
        betas: [0.5, 0.9]
        weight_decay: 0.0
    scheduler:
      type: MultiStepLR
      params:
        gamma: 0.5
        milestones:
            - 200000
            - 400000
            - 600000
            - 800000

###########################################################
#       DISCRIMINATOR NETWORK ARCHITECTURE SETTING        #
###########################################################
  MultiScaleDiscriminator:
    params:
      scales: 3                              
      downsample_pooling: "DWT"  
      downsample_pooling_params:
          kernel_size: 4                    
          stride: 2                         
          padding: 2                        
      discriminator_params:
          in_channels: 1                     
          out_channels: 1                    
          kernel_sizes: [15, 41, 5, 3]       
          channels: 128                      
          max_downsample_channels: 1024     
          max_groups: 16                   
          bias: true
          downsample_scales: [4, 4, 4, 4, 1]
          nonlinear_activation: "LeakyReLU"  
          nonlinear_activation_params:
            negative_slope: 0.1
      follow_official_norm: true    
    optimizer:
      type: Adam
      params:
        lr: 2.0e-4
        betas: [0.5, 0.9]
        weight_decay: 0.0
    scheduler:
      type: MultiStepLR
      params:
        gamma: 0.5
        milestones:
            - 200000
            - 400000
            - 600000
            - 800000

  MultiPeriodDiscriminator:
    params:
      periods: [2, 3, 5, 7, 11]      
      discriminator_params:
        in_channels: 1                  
        out_channels: 1                  
        kernel_sizes: [5, 3]              
        channels: 32                       
        downsample_scales: [3, 3, 3, 3, 1] 
        max_downsample_channels: 1024      
        bias: true                       
        nonlinear_activation: "LeakyReLU"  
        nonlinear_activation_params:       
          negative_slope: 0.1
        use_spectral_norm: false           
    optimizer:
      type: Adam
      params:
        lr: 2.0e-4
        betas: [0.5, 0.9]
        weight_decay: 0.0
    scheduler:
      type: MultiStepLR
      params:
        gamma: 0.5
        milestones:
            - 200000
            - 400000
            - 600000
            - 800000

####################################################
#                   LOSS SETTING                   #
####################################################
Loss:
  generator_adv_loss:
    enable: True
    params:
      average_by_discriminators: False
    weights: 1.0

  discriminator_adv_loss:
    enable: True
    params:
      average_by_discriminators: False
    weights: 1.0

  stft_loss:
    enable: False             # Whether to use multi-resolution STFT loss.

  mel_loss:
    enable: True
    params:
      fs: 16000
      fft_size: 2048
      hop_size: 200
      win_length: 1000
      window: "hann"
      num_mels: 80
      fmin: 0
      fmax: 8000
      log_base: null
    weights: 45.0

  subband_stft_loss:
    enable: False
    params:
      fft_sizes: [384, 683, 171]  # List of FFT size for STFT-based loss.
      hop_sizes: [35, 75, 15]     # List of hop size for STFT-based loss
      win_lengths: [150, 300, 60] # List of window length for STFT-based loss.
      window: "hann_window"       # Window function for STFT-based loss

  feat_match_loss:
    enable: True
    params:
      average_by_discriminators: false 
      average_by_layers: false         
    weights: 2.0


###########################################################
#                  DATA LOADER SETTING                    #
###########################################################
batch_size: 16              
batch_max_steps: 9600       # Length of each audio in batch. Make sure dividable by hop_size.
pin_memory: True            
num_workers: 2 # FIXME: set > 0 may stuck on macos              
remove_short_samples: False 
allow_cache: True           

generator_grad_norm: -1

discriminator_grad_norm: -1

###########################################################
#                    INTERVAL SETTING                     #
###########################################################
generator_train_start_steps: 1     # Number of steps to start to train discriminator.
discriminator_train_start_steps: 0 # Number of steps to start to train discriminator.
train_max_steps: 2500000           # Number of training steps.
save_interval_steps: 20000         # Interval steps to save checkpoint.
eval_interval_steps: 10000          # Interval steps to evaluate the network.
log_interval_steps: 1000            # Interval steps to record the training log.

###########################################################
#                     OTHER SETTING                       #
###########################################################
num_save_intermediate_results: 4  # Number of results to be saved as intermediate results.