audio:
  num_mels: 80  #the number of mel bands when calculating mel spectrograms.
  n_fft: 2048 #the number of fft components.
  sr: 22050 #the sampling rate of audio data file.
  preemphasis: 0.97 #the preemphasis coefficient.
  hop_length: 256 #the number of samples to advance between frames.
  win_length: 1024 #the length (width) of the window function.
  power: 1.2 #the power to raise before griffin-lim.
  min_level_db: -100 #the minimum level db.
  ref_level_db: 20 #the reference level db.
  outputs_per_step: 1 #the outputs per step.

encoder_n_layer: 6 #the number of FFT Block in encoder.
encoder_head: 2 #the attention head number in encoder.
encoder_conv1d_filter_size: 1536 #the filter size of conv1d in encoder.
max_seq_len: 2048 #the max length of sequence.
decoder_n_layer: 6 #the number of FFT Block in decoder.
decoder_head: 2 #the attention head number in decoder.
decoder_conv1d_filter_size: 1536 #the filter size of conv1d in decoder.
fs_hidden_size: 384 #the hidden size in model of fastspeech.
duration_predictor_output_size: 256 #the output size of duration predictior.
duration_predictor_filter_size: 3 #the filter size of conv1d in duration prediction.
fft_conv1d_filter: 3 #the filter size of conv1d in fft.
fft_conv1d_padding: 1 #the padding size of conv1d in fft.
dropout: 0.1 #the dropout in network.
transformer_head: 4 #the attention head num of transformerTTS.

embedding_size: 512 #the dim size of embedding of transformerTTS.
hidden_size: 256 #the hidden size in model of transformerTTS.
warm_up_step: 4000 #the warm up step of learning rate.
grad_clip_thresh: 0.1 #the threshold of grad clip.