JorisCos commited on
Commit
38fc890
1 Parent(s): cbb05f3

Initial commit

Browse files
Files changed (3) hide show
  1. .gitattributes +1 -0
  2. README.md +75 -0
  3. pytorch_model.bin +3 -0
.gitattributes CHANGED
@@ -6,3 +6,4 @@
6
  *.tar.gz filter=lfs diff=lfs merge=lfs -text
7
  *.ot filter=lfs diff=lfs merge=lfs -text
8
  *.onnx filter=lfs diff=lfs merge=lfs -text
 
6
  *.tar.gz filter=lfs diff=lfs merge=lfs -text
7
  *.ot filter=lfs diff=lfs merge=lfs -text
8
  *.onnx filter=lfs diff=lfs merge=lfs -text
9
+ pytorch_model.bin filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - asteroid
4
+ - audio
5
+ - ConvTasNet
6
+ datasets:
7
+ - Libri2Mix
8
+ - sep_clean
9
+ license: CC BY-NC 4.0
10
+ inference: false
11
+ ---
12
+
13
+ ## Asteroid model `JorisCos/ConvTasNet_Libri2Mix_sepclean_8k`
14
+ Imported from [Zenodo](https://zenodo.org/record/3873572#.X9M69cLjJH4)
15
+
16
+ Description:
17
+
18
+ This model was trained by Joris Cosentino using the librimix recipe in [Asteroid](https://github.com/asteroid-team/asteroid).
19
+ It was trained on the `sep_clean` task of the Libri2Mix dataset.
20
+
21
+ Training config:
22
+ ```yaml
23
+ data:
24
+ n_src: 2
25
+ sample_rate: 8000
26
+ segment: 3
27
+ task: sep_clean
28
+ train_dir: data/wav8k/min/train-360
29
+ valid_dir: data/wav8k/min/dev
30
+ filterbank:
31
+ kernel_size: 16
32
+ n_filters: 512
33
+ stride: 8
34
+ masknet:
35
+ bn_chan: 128
36
+ hid_chan: 512
37
+ mask_act: relu
38
+ n_blocks: 8
39
+ n_repeats: 3
40
+ skip_chan: 128
41
+ optim:
42
+ lr: 0.001
43
+ optimizer: adam
44
+ weight_decay: 0.0
45
+ training:
46
+ batch_size: 24
47
+ early_stop: True
48
+ epochs: 200
49
+ half_lr: True
50
+ num_workers: 2
51
+ ```
52
+
53
+
54
+ Results :
55
+
56
+ On Libri2Mix min test set :
57
+ ```yaml
58
+ si_sdr: 14.764543634468069
59
+ si_sdr_imp: 14.764029375607246
60
+ sdr: 15.29337970745095
61
+ sdr_imp: 15.114146605113111
62
+ sir: 24.092904661115366
63
+ sir_imp: 23.913669683141528
64
+ sar: 16.06055906916849
65
+ sar_imp: -51.980784441287454
66
+ stoi: 0.9311142440593033
67
+ stoi_imp: 0.21817376142710482
68
+ ```
69
+
70
+ License notice:
71
+
72
+ This work "ConvTasNet_Libri2Mix_sepclean_8k"
73
+ is a derivative of LibriSpeech ASR corpus by Vassil Panayotov,
74
+ used under CC BY 4.0. "ConvTasNet_Libri2Mix_sepclean_8k"
75
+ is licensed under Attribution-ShareAlike 3.0 Unported by Cosentino Joris.
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bf96b1e7adb377592341b965bb9c5a2da2e4a5ab8630a43e1d9cd0f20793252
3
+ size 20331472