configs/config_htdemucs_bass.yaml

audio:
  chunk_size: 485100 # samplerate * segment
  min_mean_abs: 0.001
  hop_length: 1024

training:
  batch_size: 8
  gradient_accumulation_steps: 1
  grad_clip: 0
  segment: 11
  shift: 1
  samplerate: 44100
  channels: 2
  normalize: true
  instruments: ['drums', 'bass', 'other', 'vocals']
  target_instrument: null
  num_epochs: 1000
  num_steps: 1000
  optimizer: adam
  lr: 9.0e-05
  patience: 2
  reduce_factor: 0.95
  q: 0.95
  coarse_loss_clip: true
  ema_momentum: 0.999
  other_fix: false # it's needed for checking on multisong dataset if other is actually instrumental
  use_amp: true # enable or disable usage of mixed precision (float16) - usually it must be true

augmentations:
  enable: true # enable or disable all augmentations (to fast disable if needed)
  loudness: true # randomly change loudness of each stem on the range (loudness_min; loudness_max)
  loudness_min: 0.5
  loudness_max: 1.5

inference:
  num_overlap: 4
  batch_size: 8

model: htdemucs

htdemucs:  # see demucs/htdemucs.py for a detailed description
  # Channels
  channels: 48
  channels_time:
  growth: 2
  # STFT
  num_subbands: 1
  nfft: 4096
  wiener_iters: 0
  end_iters: 0
  wiener_residual: false
  cac: true
  # Main structure
  depth: 4
  rewrite: true
  # Frequency Branch
  multi_freqs: []
  multi_freqs_depth: 3
  freq_emb: 0.2
  emb_scale: 10
  emb_smooth: true
  # Convolutions
  kernel_size: 8
  stride: 4
  time_stride: 2
  context: 1
  context_enc: 0
  # normalization
  norm_starts: 4
  norm_groups: 4
  # DConv residual branch
  dconv_mode: 3
  dconv_depth: 2
  dconv_comp: 8
  dconv_init: 1e-3
  # Before the Transformer
  bottom_channels: 512
  # CrossTransformer
  # ------ Common to all
  # Regular parameters
  t_layers: 5
  t_hidden_scale: 4.0
  t_heads: 8
  t_dropout: 0.0
  t_layer_scale: True
  t_gelu: True
  # ------------- Positional Embedding
  t_emb: sin
  t_max_positions: 10000 # for the scaled embedding
  t_max_period: 10000.0
  t_weight_pos_embed: 1.0
  t_cape_mean_normalize: True
  t_cape_augment: True
  t_cape_glob_loc_scale: [5000.0, 1.0, 1.4]
  t_sin_random_shift: 0
  # ------------- norm before a transformer encoder
  t_norm_in: True
  t_norm_in_group: False
  # ------------- norm inside the encoder
  t_group_norm: False
  t_norm_first: True
  t_norm_out: True
  # ------------- optim
  t_weight_decay: 0.0
  t_lr:
  # ------------- sparsity
  t_sparse_self_attn: False
  t_sparse_cross_attn: False
  t_mask_type: diag
  t_mask_random_seed: 42
  t_sparse_attn_window: 400
  t_global_window: 100
  t_sparsity: 0.95
  t_auto_sparsity: False
  # Cross Encoder First (False)
  t_cross_first: False
  # Weight init
  rescale: 0.1