Hugging Face's logo

crlandsc
/
bsrnn-drums

audio source separation
music demixing
band-split recurrent neural network
bsrnn
spectrogram
drums
Model card Files Community
1
bsrnn-drums
File size: 3,707 Bytes 
c685cb8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
 
model:
  sr: 44100
  n_fft: 2048
  bandsplits:
  - - 1000
    - 50
  - - 2000
    - 100
  - - 4000
    - 250
  - - 8000
    - 500
  - - 16000
    - 1000
  bottleneck_layer: rnn
  t_timesteps: 263
  fc_dim: 128
  rnn_dim: 256
  rnn_type: LSTM
  bidirectional: true
  num_layers: 10
  mlp_dim: 512
  return_mask: false
  complex_as_channel: true
  is_mono: false
train_dataset:
  file_dir: /home/afteraugustmusician/Music-Demixing-with-Band-Split-RNN/datasets/Drums
  txt_dir: files/
  txt_path: null
  target: drums
  is_training: true
  is_mono: false
  sr: 44100
  preload_dataset: false
  silent_prob: 0.1
  mix_prob: 0.25
  mix_tgt_too: false
test_dataset:
  in_fp: /home/afteraugustmusician/Music-Demixing-with-Band-Split-RNN/datasets/Drums
  target: drums
  is_mono: false
  sr: 44100
  win_size: 3
  hop_size: 0.5
  batch_size: 4
  window: null
sad:
  sr: 44100
  window_size_in_sec: 6
  overlap_ratio: 0.5
  n_chunks_per_segment: 10
  eps: 1.0e-05
  gamma: 0.001
  threshold_max_quantile: 0.15
  threshold_segment: 0.5
augmentations:
  randomcrop:
    _target_: data.augmentations.RandomCrop
    p: 1
    chunk_size_sec: 3
    sr: 44100
    window_stft: 2048
    hop_stft: 512
  gainscale:
    _target_: data.augmentations.GainScale
    p: 0.5
    min_db: -10.0
    max_db: 10.0
featurizer:
  direct_transform:
    _target_: torchaudio.transforms.Spectrogram
    n_fft: 2048
    win_length: 2048
    hop_length: 512
    power: null
  inverse_transform:
    _target_: torchaudio.transforms.InverseSpectrogram
    n_fft: 2048
    win_length: 2048
    hop_length: 512
callbacks:
  lr_monitor:
    _target_: pytorch_lightning.callbacks.LearningRateMonitor
    logging_interval: epoch
  model_ckpt:
    _target_: pytorch_lightning.callbacks.ModelCheckpoint
    monitor: train/loss
    mode: min
    save_top_k: 5
    dirpath: /home/afteraugustmusician/Music-Demixing-with-Band-Split-RNN/src/logs/bandsplitrnn/2023-05-08_15-35/weights
    filename: epoch{epoch:02d}-train_loss{train/loss:.2f}
    auto_insert_metric_name: false
  model_ckpt_usdr:
    _target_: pytorch_lightning.callbacks.ModelCheckpoint
    monitor: train/usdr
    mode: max
    save_top_k: 5
    dirpath: /home/afteraugustmusician/Music-Demixing-with-Band-Split-RNN/src/logs/bandsplitrnn/2023-05-08_15-35/weights
    filename: epoch{epoch:02d}-train_usdr{train/usdr:.2f}
    auto_insert_metric_name: false
  ema:
    _target_: utils.callbacks.EMA
    decay: 0.9999
    validate_original_weights: false
    every_n_steps: 1
logger:
  tensorboard:
    _target_: pytorch_lightning.loggers.TensorBoardLogger
    save_dir: /home/afteraugustmusician/Music-Demixing-with-Band-Split-RNN/src/logs/bandsplitrnn/2023-05-08_15-35/tb_logs
    name: ''
    version: ''
    log_graph: false
    default_hp_metric: false
    prefix: ''
  wandb:
    _target_: pytorch_lightning.loggers.WandbLogger
    project: MDX_BSRNN_23
    name: drums
    save_dir: wandb_logs
    offline: false
    id: null
    log_model: false
    prefix: ''
    job_type: train
    group: ''
    tags: []
train_loader:
  batch_size: 8
  num_workers: 12
  shuffle: true
  drop_last: true
val_loader:
  batch_size: 2
  num_workers: 8
  shuffle: false
  drop_last: false
opt:
  _target_: torch.optim.Adam
  lr: 0.001
sch:
  warmup_step: 10
  alpha: 1
  gamma: 0.9899494936611665
ckpt_path: logs/bandsplitrnn/2023-04-29_14-45/weights/drums-193-usdr-5.29.ckpt
trainer:
  fast_dev_run: false
  min_epochs: 100
  max_epochs: 500
  log_every_n_steps: 10
  accelerator: auto
  devices: auto
  gradient_clip_val: 5
  precision: 32
  enable_progress_bar: true
  benchmark: true
  deterministic: false
experiment_dirname: bandsplitrnn
wandb_api_key: d5c4447e39b2b10b95f05f907d57845ded16bc13