groadabike
commited on
Commit
·
9ee95da
1
Parent(s):
dc36d06
Model release: v1
Browse files- README.md +105 -0
- pytorch.bin +3 -0
README.md
ADDED
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
tags:
|
3 |
+
- asteroid
|
4 |
+
- audio
|
5 |
+
- ConvTasNet
|
6 |
+
datasets:
|
7 |
+
- DAMP-VSEP (English+Non-English training set)
|
8 |
+
- Singing/Accompaniment Separation
|
9 |
+
license: Restricted License
|
10 |
+
---
|
11 |
+
|
12 |
+
|
13 |
+
## Description:
|
14 |
+
This model was trained by Gerardo Roa using the dampvsep recipe in Asteroid.
|
15 |
+
It was trained on the singing/accompaniment task of the DAMP-VSEP dataset.
|
16 |
+
|
17 |
+
|
18 |
+
## Training config:
|
19 |
+
```yaml
|
20 |
+
data:
|
21 |
+
channels: 1
|
22 |
+
emb_model: 'no'
|
23 |
+
metadata_path: metadata
|
24 |
+
mixture: remix
|
25 |
+
root_path: /fastdata/acp13gr/DAMP/DAMP-VSEP
|
26 |
+
sample_rate: 16000
|
27 |
+
train_set: english_nonenglish
|
28 |
+
filterbank:
|
29 |
+
kernel_size: 20
|
30 |
+
n_filters: 256
|
31 |
+
stride: 10
|
32 |
+
main_args:
|
33 |
+
exp_dir: exp/train_convtasnet_remix-no-0.0-english_nonenglish-0.0005-jade
|
34 |
+
help: null
|
35 |
+
masknet:
|
36 |
+
bn_chan: 256
|
37 |
+
conv_kernel_size: 3
|
38 |
+
hid_chan: 512
|
39 |
+
mask_act: relu
|
40 |
+
n_blocks: 10
|
41 |
+
n_repeats: 4
|
42 |
+
n_src: 2
|
43 |
+
norm_type: gLN
|
44 |
+
skip_chan: 256
|
45 |
+
optim:
|
46 |
+
lr: 0.0005
|
47 |
+
optimizer: adam
|
48 |
+
weight_decay: 0.0
|
49 |
+
positional arguments: {}
|
50 |
+
training:
|
51 |
+
batch_size: 7
|
52 |
+
early_stop: true
|
53 |
+
epochs: 50
|
54 |
+
half_lr: true
|
55 |
+
loss_alpha: 0.0
|
56 |
+
num_workers: 10
|
57 |
+
```
|
58 |
+
|
59 |
+
|
60 |
+
## Results:
|
61 |
+
```yaml
|
62 |
+
"si_sdr": 15.111802516750586,
|
63 |
+
"si_sdr_imp": 15.178209807687663,
|
64 |
+
"si_sdr_s0": 12.160261214703553,
|
65 |
+
"si_sdr_s0_imp": 17.434593619085675,
|
66 |
+
"si_sdr_s1": 18.063343818797623,
|
67 |
+
"si_sdr_s1_imp": 12.92182599628965,
|
68 |
+
"sdr": 15.959722569460281,
|
69 |
+
"sdr_imp": 14.927002467087567,
|
70 |
+
"sdr_s0": 13.270412028426595,
|
71 |
+
"sdr_s0_imp": 16.45867572657551,
|
72 |
+
"sdr_s1": 18.64903311049397,
|
73 |
+
"sdr_s1_imp": 13.39532920759962,
|
74 |
+
"sir": 23.935932341084754,
|
75 |
+
"sir_imp": 22.903212238712012,
|
76 |
+
"sir_s0": 22.30777879911744,
|
77 |
+
"sir_s0_imp": 25.49604249726635,
|
78 |
+
"sir_s1": 25.56408588305207,
|
79 |
+
"sir_s1_imp": 20.310381980157665,
|
80 |
+
"sar": 17.174899162445882,
|
81 |
+
"sar_imp": -134.47377304178818,
|
82 |
+
"sar_s0": 14.268071153965913,
|
83 |
+
"sar_s0_imp": -137.38060105026818,
|
84 |
+
"sar_s1": 20.081727170925856,
|
85 |
+
"sar_s1_imp": -131.56694503330817,
|
86 |
+
"stoi": 0.7746496376326059,
|
87 |
+
"stoi_imp": 0.19613735629114643,
|
88 |
+
"stoi_s0": 0.6611376621212413,
|
89 |
+
"stoi_s0_imp": 0.21162695175464794,
|
90 |
+
"stoi_s1": 0.8881616131439705,
|
91 |
+
"stoi_s1_imp": 0.1806477608276449
|
92 |
+
```
|
93 |
+
|
94 |
+
|
95 |
+
## License notice:
|
96 |
+
|
97 |
+
** This is important, please fill it, if you need help, you can ask on Asteroid's slack.**
|
98 |
+
|
99 |
+
This work "ConvTasNet_DAMPVSEP_EnglishNonEnglish_baseline"
|
100 |
+
is a derivative of [DAMP-VSEP](https://zenodo.org/record/3553059) by
|
101 |
+
[Smule, Inc](https://www.smule.com/),
|
102 |
+
used under [Reastricte License](https://zenodo.org/record/3553059).
|
103 |
+
"ConvTasNet_DAMPVSEP_EnglishNonEnglish_baseline"
|
104 |
+
is licensed under [Attribution-ShareAlike 3.0 Unported](https://creativecommons.org/licenses/by-sa/3.0/)
|
105 |
+
by Gerardo Roa.
|
pytorch.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f77ed26005b8cc6d9b6ca4f313e252b4b80b17378a0097c47eb60811708b75b0
|
3 |
+
size 64766287
|