groadabike commited on
Commit
c95ee17
1 Parent(s): 5e1587f

Initial commit

Browse files
Files changed (2) hide show
  1. README.md +77 -0
  2. pytorch_model.bin +3 -0
README.md ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - asteroid
4
+ - audio
5
+ - ConvTasNet
6
+ - audio-source-separation
7
+ datasets:
8
+ - wham
9
+ - sep_clean
10
+ license: cc-by-sa-3.0
11
+ inference: false
12
+ ---
13
+
14
+ ## Asteroid model `mpariente/ConvTasNet_WHAM_sepclean`
15
+ Imported from [Zenodo](https://zenodo.org/record/3862942)
16
+
17
+ ### Description:
18
+ This model was trained by Manuel Pariente
19
+ using the wham/ConvTasNet recipe in [Asteroid](https://github.com/asteroid-team/asteroid).
20
+ It was trained on the `sep_clean` task of the WHAM! dataset.
21
+
22
+ ### Training config:
23
+ ```yaml
24
+ data:
25
+ channels: 1
26
+ n_src: 2
27
+ root_path: data
28
+ sample_rate: 16000
29
+ samples_per_track: 10
30
+ segment: 3.0
31
+ task: enh_both
32
+ filterbank:
33
+ kernel_size: 20
34
+ n_filters: 256
35
+ stride: 10
36
+ main_args:
37
+ exp_dir: exp/train_convtasnet
38
+ help: None
39
+ masknet:
40
+ bn_chan: 256
41
+ conv_kernel_size: 3
42
+ hid_chan: 512
43
+ mask_act: relu
44
+ n_blocks: 8
45
+ n_repeats: 4
46
+ n_src: 2
47
+ norm_type: gLN
48
+ skip_chan: 256
49
+ optim:
50
+ lr: 0.0003
51
+ optimizer: adam
52
+ weight_decay: 0.0
53
+ positional arguments:
54
+ training:
55
+ batch_size: 12
56
+ early_stop: True
57
+ epochs: 50
58
+ half_lr: True
59
+ num_workers: 12
60
+ ```
61
+
62
+ ### Results:
63
+ ```yaml
64
+ si_sdr: 14.018196157142519
65
+ si_sdr_imp: 14.017103133809577
66
+ sdr: 14.498517291333885
67
+ sdr_imp: 14.463389151567865
68
+ sir: 24.149634529133372
69
+ sir_imp: 24.11450638936735
70
+ sar: 15.338597389045935
71
+ sar_imp: -137.30634122401517
72
+ stoi: 0.7639416744417206
73
+ stoi_imp: 0.1843383526963759
74
+ ```
75
+
76
+ ### License notice:
77
+ This work "ConvTasNet_DAMP-VSEP_enhboth" is a derivative of DAMP-VSEP: Smule Digital Archive of Mobile Performances - Vocal Separation (Version 1.0.1) by Smule, Inc, used under Smule's Research Data License Agreement (Research only). "ConvTasNet_DAMP-VSEP_enhboth" is licensed under Attribution-ShareAlike 3.0 Unported by Gerardo Roa Dabike.
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8519e8658572f0d3a5e07002849337cb0ff07dcf3b3a641244e0905ceb0adc44
3
+ size 51990656