npc_example.yml

data:
  # Dataset-related setting
  dataset:
    name: 'LibriSpeech'    # Specify dataset, must match to dataset/<corpus>.py
    path: '/path/to/LibriSpeech'         # Path to unzipped LibriSpeech dataset
    train_split: ['train-clean-360']        # Splits to be used as training set
    dev_split: ['dev-clean']                # Splits to be used as valid.   set
    batch_size: 32                                                # Batch sizes
    audio_max_frames: 1500     # Max length of spectrogram to ensure batch size
  # Attributes of audio feature
  audio:
    feat_type: 'fbank'                                           # Feature type
    feat_dim:  80                                           # Feature dimension
    frame_length: 25                                        # Window size in ms
    frame_shift: 10                                            # Hop size in ms
    cmvn: True                      # Apply uttr.-wised CMVN on Mel spectrogram

model:
  method: 'npc'                                         # Accepts npc/apc/vqapc
  paras:
    kernel_size: 15     # Receptive field size (R) = kernel_size + 2*(n_blocks)
    mask_size: 5     # Desired input mask size (M_in) as described in NPC paper
    n_blocks: 4                     # Number of ConvBlocks stacked in NPC model
    hidden_size: 512                       # Dimension of feature of all layers
    dropout: 0.1                                         # Dropout in ConvBlock
    residual: True                           # Residual connection in ConvBlock
    batch_norm: True                             # Apply BatchNorm in ConvBlock
    activate: 'relu'                         # Activation function of ConvBlock
    disable_cross_layer: False      # Apply Masked ConvBlock at last layer only
    vq:
      codebook_size: [64,64,64,64]    # Codebook size of each group in VQ-layer
      code_dim: [128,128,128,128] # Dim of each group summing up to hidden_size
      gumbel_temperature: 1.0       # Temperature of Gumbel Softmax in VQ-layer

hparas:
    optimizer: 'Adam'
    lr: 0.001
    epoch: 100