Upload 6 files

MDX23C-DrumSep-aufr33-jarredou.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2a4aa53eb584d21eead358a4e66d1882ad182911be018f052b5da73be9096d0
+size 437652699

config_bs_roformer_sw.yaml

@@ -0,0 +1,197 @@
+audio:
+  chunk_size: 588800 #882000
+  dim_f: 1024
+  dim_t: 801 # don't work (use in model)
+  hop_length: 441 # don't work (use in model)
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.000
+
+model:
+  dim: 256
+  depth: 12
+  stereo: true
+  num_stems: 6
+  time_transformer_depth: 1
+  freq_transformer_depth: 1
+  linear_transformer_depth: 0
+  freqs_per_bands: !!python/tuple
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 2
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 4
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 12
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 24
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 48
+    - 128
+    - 129
+  dim_head: 64
+  heads: 8
+  attn_dropout: 0.1
+  ff_dropout: 0.1
+  flash_attn: true
+  dim_freqs_in: 1025
+  stft_n_fft: 2048
+  stft_hop_length: 512
+  stft_win_length: 2048
+  stft_normalized: false
+  mask_estimator_depth: 2
+  multi_stft_resolution_loss_weight: 1.0
+  multi_stft_resolutions_window_sizes: !!python/tuple
+  - 4096
+  - 2048
+  - 1024
+  - 512
+  - 256
+  multi_stft_hop_size: 147
+  multi_stft_normalized: False
+  mlp_expansion_factor: 4
+  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
+
+training:
+  batch_size: 2
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments: ['bass', 'drums', 'other', 'vocals', 'guitar', 'piano']
+  patience: 3
+  reduce_factor: 0.95
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1000
+  augmentation: false # enable augmentations by audiomentations and pedalboard
+  augmentation_type: simple1
+  use_mp3_compress: false # Deprecated
+  augmentation_mix: true # Mix several stems of the same type with some probability
+  augmentation_loudness: true # randomly change loudness of each stem
+  augmentation_loudness_type: 1 # Type 1 or 2
+  augmentation_loudness_min: 0.5
+  augmentation_loudness_max: 1.5
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  # optimizer: prodigy
+  optimizer: adam
+  # lr: 1.0
+  lr: 1.0e-5
+  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: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02)
+    - 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)
+
+  vocals:
+      pitch_shift: 0.1
+      pitch_shift_min_semitones: -5
+      pitch_shift_max_semitones: 5
+      seven_band_parametric_eq: 0.1
+      seven_band_parametric_eq_min_gain_db: -9
+      seven_band_parametric_eq_max_gain_db: 9
+      tanh_distortion: 0.1
+      tanh_distortion_min: 0.1
+      tanh_distortion_max: 0.7
+  bass:
+    pitch_shift: 0.1
+    pitch_shift_min_semitones: -2
+    pitch_shift_max_semitones: 2
+    seven_band_parametric_eq: 0.1
+    seven_band_parametric_eq_min_gain_db: -3
+    seven_band_parametric_eq_max_gain_db: 6
+    tanh_distortion: 0.1
+    tanh_distortion_min: 0.1
+    tanh_distortion_max: 0.5
+  drums:
+    pitch_shift: 0.1
+    pitch_shift_min_semitones: -5
+    pitch_shift_max_semitones: 5
+    seven_band_parametric_eq: 0.1
+    seven_band_parametric_eq_min_gain_db: -9
+    seven_band_parametric_eq_max_gain_db: 9
+    tanh_distortion: 0.1
+    tanh_distortion_min: 0.1
+    tanh_distortion_max: 0.6
+  other:
+    pitch_shift: 0.1
+    pitch_shift_min_semitones: -4
+    pitch_shift_max_semitones: 4
+    gaussian_noise: 0.1
+    gaussian_noise_min_amplitude: 0.001
+    gaussian_noise_max_amplitude: 0.015
+    time_stretch: 0.1
+    time_stretch_min_rate: 0.8
+    time_stretch_max_rate: 1.25
+
+
+inference:
+  batch_size: 1
+  dim_t: 1101
+  num_overlap: 2
+  normalize: false

config_drumsep_5stems_mdx23c.yaml

@@ -0,0 +1,70 @@
+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: 0.000
+
+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 # 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
+
+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: 2
+  dim_t: 512
+  num_overlap: 4
+  normalize: false

config_drumsep_mdx23c.yaml

@@ -0,0 +1,87 @@
+audio:
+  chunk_size: 130560
+  dim_f: 1024
+  dim_t: 256
+  hop_length: 512
+  n_fft: 2048
+  num_channels: 2
+  sample_rate: 44100
+  min_mean_abs: 0.001
+
+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: 12
+  gradient_accumulation_steps: 1
+  grad_clip: 0
+  instruments:
+  - kick
+  - snare
+  - toms
+  - hh
+  - ride
+  - crash
+  lr: 9.0e-05
+  patience: 30
+  reduce_factor: 0.95
+  target_instrument: null
+  num_epochs: 1000
+  num_steps: 1268
+  q: 0.95
+  coarse_loss_clip: true
+  ema_momentum: 0.999
+  optimizer: adam
+  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: !!python/tuple # 2 additional stems of the same type (1st with prob 0.2, 2nd with prob 0.02)
+    - 0.2
+    - 0.02
+  mixup_loudness_min: 0.5
+  mixup_loudness_max: 1.5
+
+  # apply mp3 compression to mixture only (emulate downloading mp3 from internet)
+  mp3_compression_on_mixture: 0.0
+  mp3_compression_on_mixture_bitrate_min: 32
+  mp3_compression_on_mixture_bitrate_max: 320
+  mp3_compression_on_mixture_backend: "lameenc"
+
+  all:
+    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)
+    mp3_compression: 0.0
+    mp3_compression_min_bitrate: 32
+    mp3_compression_max_bitrate: 320
+    mp3_compression_backend: "lameenc"
+    pitch_shift: 0.1
+    pitch_shift_min_semitones: -3
+    pitch_shift_max_semitones: 3
+    seven_band_parametric_eq: 0.5
+    seven_band_parametric_eq_min_gain_db: -6
+    seven_band_parametric_eq_max_gain_db: 6
+    tanh_distortion: 0.2
+    tanh_distortion_min: 0.1
+    tanh_distortion_max: 0.5
+    
+inference:
+  batch_size: 1
+  dim_t: 256
+  num_overlap: 4

drumsep_5stems_mdx23c_jarredou.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f8e636fb674b88a52c8399fde9a4ebe2b72b065ca07eed4e03ab1c9f0bfb2e0
+size 437648758

model_bs_roformer_sw.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24e7d35ee9c64415673d3fd33e06a67cac2c103c5df6267ba1576459c775916e
+size 699412152