audio:
  chunk_size: 260096
  dim_f: 6144
  dim_t: 128
  hop_length: 2048
  n_fft: 12288
  num_channels: 2
  sample_rate: 44100
  min_mean_abs: 0.001
model:
  act: gelu
  bottleneck_factor: 4
  growth: 64
  norm: InstanceNorm
  num_blocks_per_scale: 2
  num_channels: 128
  num_scales: 5
  num_subbands: 6
  scale:
  - 2
  - 2
training:
  batch_size: 14
  grad_clip: 0
  instruments:
  - Vocals
  - Instrumental
  lr: 3.0e-05
  target_instrument: null
  num_epochs: 1000
  num_steps: 1000
  augmentation: 1
inference:
  batch_size: 1
  dim_t: 256
  num_overlap: 8