Add files using upload-large-folder tool

htdemucs/demucs3_mmi_config.yaml

@@ -0,0 +1,72 @@
+audio:
+  chunk_size: 485100 # samplerate * segment
+  min_mean_abs: 0.000
+  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: false # 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: hdemucs
+
+hdemucs:  # see demucs/hdemucs.py for a detailed description
+  channels: 48
+  channels_time: null
+  growth: 2
+  nfft: 4096
+  wiener_iters: 0
+  end_iters: 0
+  wiener_residual: False
+  cac: True
+  depth: 6
+  rewrite: True
+  hybrid: True
+  hybrid_old: False
+  multi_freqs: []
+  multi_freqs_depth: 3
+  freq_emb: 0.2
+  emb_scale: 10
+  emb_smooth: True
+  kernel_size: 8
+  stride: 4
+  time_stride: 2
+  context: 1
+  context_enc: 0
+  norm_starts: 4
+  norm_groups: 4
+  dconv_mode: 1
+  dconv_depth: 2
+  dconv_comp: 4
+  dconv_attn: 4
+  dconv_lstm: 4
+  dconv_init: 0.001
+  rescale: 0.1

htdemucs/demucs4_4stem_config.yaml

@@ -0,0 +1,119 @@
+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
+

htdemucs/demucs4_6stem_config.yaml

@@ -0,0 +1,127 @@
+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', 'guitar', 'piano']
+  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
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: [0.2, 0.02]
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+  all:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.1 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+
+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: 0
+  # 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
+

htdemucs/demucs4_choirsep_config.yaml

@@ -0,0 +1,151 @@
+audio:
+  chunk_size: 132300 # samplerate * segment
+  min_mean_abs: 0.001
+  hop_length: 1024
+
+training:
+  batch_size: 4
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  segment: 3
+  shift: 1
+  samplerate: 44100
+  channels: 2
+  normalize: true
+  instruments: ['alto', 'bass', 'soprano', 'tenor']
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1000
+  optimizer: adam
+  lr: 1.0e-04
+  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: false # enable or disable usage of mixed precision (float16) - usually it must be true
+
+loss_multistft:
+  fft_sizes:
+  - 1024
+  - 2048
+  - 4096
+  hop_sizes:
+  - 512
+  - 1024
+  - 2048
+  win_lengths:
+  - 1024
+  - 2048
+  - 4096
+  window: "hann_window"
+  scale: "mel"
+  n_bins: 128
+  sample_rate: 44100
+  perceptual_weighting: true
+  w_sc: 1.0
+  w_log_mag: 1.0
+  w_lin_mag: 0.0
+  w_phs: 0.0
+  mag_distance: "L1"
+
+augmentations:
+  enable: false # 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
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: [0.2, 0.02]
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+  all:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.1 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+
+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: 0
+  # 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
+

htdemucs/demucs4_ft_bass_config.yaml

@@ -0,0 +1,119 @@
+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
+

htdemucs/demucs4_ft_drums_config.yaml

@@ -0,0 +1,119 @@
+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
+

htdemucs/demucs4_ft_other_config.yaml

@@ -0,0 +1,119 @@
+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
+

htdemucs/demucs4_ft_vocals_config.yaml

@@ -0,0 +1,119 @@
+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
+

htdemucs/demucs4_mvsep_vocals_config.yaml

@@ -0,0 +1,123 @@
+audio:
+  chunk_size: 485100 # samplerate * segment
+  min_mean_abs: 0.001
+  hop_length: 1024
+
+training:
+  batch_size: 10
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  segment: 11
+  shift: 1
+  samplerate: 44100
+  channels: 2
+  normalize: true
+  instruments: ['vocals', 'other']
+  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: true # 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
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: [0.2, 0.02]
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+
+inference:
+  num_overlap: 2
+  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
+

htdemucs/demucs_mid_side_wesleyr36_config.yaml

@@ -0,0 +1,144 @@
+audio:
+  chunk_size: 132300 # samplerate * segment
+  min_mean_abs: 0.001
+  hop_length: 1024
+
+training:
+  batch_size: 5
+  gradient_accumulation_steps: 3
+  grad_clip: 0
+  segment: 3
+  shift: 1
+  samplerate: 44100
+  channels: 2
+  normalize: true
+  instruments: ['similarity', 'difference']
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1000
+  optimizer: prodigy
+  lr: 1.0
+  patience: 80
+  reduce_factor: 0.95
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  other_fix: true # it's needed for checking on multisong dataset if other is actually instrumental
+  use_amp: false # enable or disable usage of mixed precision (float16) - usually it must be true
+
+augmentations:
+  enable: false # 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
+  mixup: true # mix several stems of same type with some probability (only works for dataset types: 1, 2, 3)
+  mixup_probs: [0.2, 0.02]
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+
+inference:
+  num_overlap: 4
+  batch_size: 18
+
+loss_multistft:
+  fft_sizes:
+  - 2048
+  - 4096
+  hop_sizes:
+  - 1024
+  - 2048
+  win_lengths:
+  - 2048
+  - 4096
+  window: "hann_window"
+  scale: "mel"
+  n_bins: 128
+  sample_rate: 44100
+  perceptual_weighting: true
+  w_sc: 1.0
+  w_log_mag: 1.0
+  w_lin_mag: 0.0
+  w_phs: 0.0
+  mag_distance: "L1"
+
+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
+

mdx23c/mdx23c_drumsep_5stem_aufr33_jarredou_config.yaml

@@ -0,0 +1,66 @@
+audio:
+  chunk_size: 523776
+  dim_f: 1024
+  dim_t: 1024
+  hop_length: 512
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.0
+model:
+  act: gelu
+  bottleneck_factor: 4
+  growth: 128
+  norm: InstanceNorm
+  num_blocks_per_scale: 2
+  num_channels: 128
+  num_scales: 5
+  num_subbands: 4
+  scale:
+    - 2
+    - 2
+training:
+  batch_size: 2
+  gradient_accumulation_steps: 4
+  grad_clip: 0
+  instruments:
+    - kick
+    - snare
+    - toms
+    - hh
+    - cymbals
+  lr: 1.0
+  patience: 200
+  reduce_factor: 0.95
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1533
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: prodigy
+  other_fix: false
+  use_amp: true
+loss_multistft:
+  fft_sizes:
+    - 2048
+  hop_sizes:
+    - 1024
+  win_lengths:
+    - 2048
+  window: hann_window
+  sample_rate: 44100
+  perceptual_weighting: false
+  scale: mel
+  n_bins: 128
+  w_sc: 0.0
+  w_log_mag: 1.0
+  w_lin_mag: 0.0
+  w_phs: 0.0
+  mag_distance: L1
+inference:
+  extension: flac
+  batch_size: 1
+  dim_t: 512
+  num_overlap: 2
+  normalize: false

mdxnet/mdx_1_9703_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.03
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_2_9682_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_3_9662_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_9482_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_crowd_hq1_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2560
+  dim_t: 256
+  n_fft: 5120
+  hop_length: 1024
+  compensation: 1.02
+  primary_stem: other
+training:
+  target_instrument: other
+  instruments:
+    - other
+    - crowd
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst1_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.045
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst2_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst3_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.028
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_187_beta_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_82_beta_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_90_beta_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_full_292_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_hq1_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_hq2_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.033
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_hq3_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.022
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_hq4_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2560
+  dim_t: 256
+  n_fft: 5120
+  hop_length: 1024
+  compensation: 1.019
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_hq5_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2560
+  dim_t: 256
+  n_fft: 5120
+  hop_length: 1024
+  compensation: 1.01
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_inst_main_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 5120
+  hop_length: 1024
+  compensation: 1.025
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_karaoke1_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_karaoke2_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 5120
+  hop_length: 1024
+  compensation: 1.065
+  primary_stem: other
+training:
+  target_instrument: other
+  instruments:
+    - other
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kim_inst_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.02
+  primary_stem: instrumental
+training:
+  target_instrument: instrumental
+  instruments:
+    - instrumental
+    - vocals
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kim_vocal1_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.043
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kim_vocal2_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.009
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_a_bass_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 512
+  n_fft: 16384
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: bass
+training:
+  target_instrument: bass
+  instruments:
+    - bass
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_a_drums_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 512
+  n_fft: 4096
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: drums
+training:
+  target_instrument: drums
+  instruments:
+    - drums
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_a_other_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 512
+  n_fft: 8192
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: other
+training:
+  target_instrument: other
+  instruments:
+    - other
+    - no_other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_a_vocals_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 512
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_b_bass_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 16384
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: bass
+training:
+  target_instrument: bass
+  instruments:
+    - bass
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_b_drums_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 128
+  n_fft: 4096
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: drums
+training:
+  target_instrument: drums
+  instruments:
+    - drums
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_b_other_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 8192
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: other
+training:
+  target_instrument: other
+  instruments:
+    - other
+    - no_other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_kuielab_b_vocals_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 2048
+  dim_t: 256
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_main_340_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.03
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_main_390_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.75
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_main_406_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.075
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_main_427_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.03
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_main_438_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.03
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_main_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.043
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_reverb_hq_foxjoy_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 512
+  n_fft: 6144
+  hop_length: 1024
+  compensation: 1.035
+  primary_stem: reverb
+training:
+  target_instrument: reverb
+  instruments:
+    - reverb
+    - other
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100

mdxnet/mdx_vocft_config.yaml

@@ -0,0 +1,20 @@
+model:
+  dim_f: 3072
+  dim_t: 256
+  n_fft: 7680
+  hop_length: 1024
+  compensation: 1.021
+  primary_stem: vocals
+training:
+  target_instrument: vocals
+  instruments:
+    - vocals
+    - instrumental
+  use_amp: true
+inference:
+  batch_size: 1
+  num_overlap: 2
+  denoise: false
+audio:
+  chunk_size: 441000
+  sample_rate: 44100