Upload config_mel_band_roformer_Lead_Rhythm_Guitar.yaml

misc/config_mel_band_roformer_Lead_Rhythm_Guitar.yaml

@@ -0,0 +1,108 @@
+audio:
+  chunk_size: 132300
+  dim_f: 1024
+  dim_t: 256
+  hop_length: 441
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 000
+
+model:
+  dim: 384
+  depth: 4
+  stereo: true
+  num_stems: 1
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  num_bands: 60
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0
+  ff_dropout: 0
+  flash_attn: true
+  dim_freqs_in: 1025
+  sample_rate: 44100  # needed for mel filter bank from librosa
+  stft_n_fft: 2048
+  stft_hop_length: 441
+  stft_win_length: 2048
+  stft_normalized: false
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 2.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+    - 4096
+    - 2048
+    - 1024
+    - 512
+    - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: false
+  mlp_expansion_factor: 2 # Probably too big (requires a lot of memory for weights)
+  use_torch_checkpoint: false  # it allows to greatly reduce GPU memory consumption during training (not fully tested)
+  skip_connection: false # Enable skip connection between transformer blocks - can solve problem with gradients and probably faster training
+
+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"
+
+training:
+  batch_size: 2
+  gradient_accumulation_steps: 2
+  grad_clip: 0
+  instruments:
+    - Lead
+    - Rhythm
+  lr: 1.0e-04
+  patience: 5
+  reduce_factor: 0.95
+  target_instrument: Lead
+  num_epochs: 1000
+  num_steps: 1000
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adamw
+  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
+  difference:
+    channel_shuffle: 0.5 # Set 0 or lower to disable
+    random_inverse: 0.01 # inverse track (better lower probability)
+    random_polarity: 0.5 # polarity change (multiply waveform to -1)
+
+inference:
+  batch_size: 12
+  dim_t: 256
+  num_overlap: 1
+
+lora:
+  r: 8
+  lora_alpha: 16. #alpha / rank > 1
+  lora_dropout: 0.05
+  merge_weights: true
+  fan_in_fan_out: false