audio:
  chunk_size: 261120
  dim_f: 1024
  dim_t: 256
  hop_length: 1024
  min_mean_abs: 0.01
  n_fft: 2048
  num_channels: 2
  sample_rate: 44100
  stereo_prob: 0.7
model:
  act: gelu
  bottleneck_factor: 4
  growth: 64
  norm: InstanceNorm
  num_blocks_per_scale: 2
  num_channels: 64
  num_scales: 5
  num_subbands: 4
  scale:
  - 2
  - 2
training:
  batch_size: 8
  ema_momentum: 0.999
  grad_clip: null
  instruments:
  - Music
  - Speech
  - SFX
  lr: 0.0001
  num_steps: 30000
  target_instrument: null
inference:
  batch_size: 8
  dim_t: 256
  instruments:
  - Music
  - Dialog
  - Effect
  num_overlap: 8