encoder:
  backbone: efficientnet_b0
  embedding_dim: 1000
  pretrained: true
feature_extractor:
  hop_length: 512
  n_fft: 2048
  spec_layer: melspectogram