Hugging Face's logo

b-sigpro
/
neural-fcasa

English
Model card Files Community
neural-fcasa
File size: 2,342 Bytes 
6d5b0af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
base_config: ${base_config_path}/train_ddp.yaml

sr: 16000
n_fft: 512
hop_length: 160

n_mic: 8
n_src: 6
dim_latent: 64

autocast: True

trainer:
  max_epochs: 200
  gradient_clip_val: 5.0
  sync_batchnorm: True

  precision: 16-mixed
  
  benchmark: True
  use_distributed_sampler: False

  callbacks:
    - _target_: lightning.pytorch.callbacks.ModelCheckpoint
      filename: "{epoch:04d}"
      save_last: True
      save_top_k: -1

    - _target_: lightning.pytorch.callbacks.RichProgressBar
      refresh_rate: 5

    - _target_: lightning.pytorch.callbacks.RichModelSummary
      max_depth: 3

    - _target_: neural_fcasa.callbacks.cyclic_annealer.CyclicAnnealerCallback
      name: beta
      cycle: 10
      max_value: 1.0
      ini_period: 50
      ini_max_value: 1.0

    - _target_: neural_fcasa.callbacks.visualizer.VisualizerCallback

dataset_name: chunk.derev-hop${hop_length}
datamodule:
  _target_: neural_fcasa.datamodules.hdf5_wavact_datamodule.DataModule
  train_dataset_path: ${working_directory}/../../hdf5/${dataset_name}-tr.hdf5
  val_dataset_path: ${working_directory}/../../hdf5/${dataset_name}-cv.hdf5
  batch_size: 2
  duration: 10
  sr: ${sr}
  hop_length: ${hop_length}

task:
  _target_: neural_fcasa.tasks.avi_scl_allsort_task.AVITask
  n_fft: ${n_fft}
  hop_length: ${hop_length}
  n_src: ${n_src}
  beta: 1.0
  gamma: 1.0

  encoder:
    _target_: neural_fcasa.encoders.resepformer_encoder.RESepFormerEncoder
    n_fft: ${n_fft}
    n_mic: ${n_mic}
    n_src: ${n_src}
    dim_latent: ${dim_latent}
    chunk_size: 100
    step_size: 100
    d_model: 256
    dim_feedforward: 1024
    n_blocks: 8
    norm_first: True
    autocast: ${autocast}

    diagonalizer:
      _target_: neural_fcasa.diagonalizers.iss_nrmxt_diagonalizer.ISSDiagonalizer
      eps: 1.e-4
      n_iter: 2

    spec_aug:
      _target_: torchaudio.transforms.SpecAugment
      n_time_masks: 2
      time_mask_param: 64
      n_freq_masks: 2
      freq_mask_param: 20
      zero_masking: True

  decoder:
    _target_: neural_fcasa.decoders.res_lin_decoder.Decoder
    n_fft: ${n_fft}
    dim_latent: ${dim_latent}
    dim_latent_noi: 10

  optimizer_config:
    _target_: aiaccel.torch.lightning.OptimizerConfig
    optimizer_generator:
      _partial_: True
      _target_: torch.optim.AdamW
      lr: 1.e-4
      weight_decay: 1.e-5