Initial commit
Browse files- README.md +65 -0
- pytorch_model.bin +3 -0
README.md
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
### Description:
|
2 |
+
This model was trained by Manuel Pariente
|
3 |
+
using the wham/ConvTasNet recipe in [Asteroid](https://github.com/asteroid-team/asteroid).
|
4 |
+
It was trained on the `sep_clean` task of the WHAM! dataset.
|
5 |
+
|
6 |
+
### Training config:
|
7 |
+
```yaml
|
8 |
+
data:
|
9 |
+
n_src: 2
|
10 |
+
mode: min
|
11 |
+
nondefault_nsrc: None
|
12 |
+
sample_rate: 8000
|
13 |
+
segment: 3
|
14 |
+
task: sep_clean
|
15 |
+
train_dir: data/wav8k/min/tr/
|
16 |
+
valid_dir: data/wav8k/min/cv/
|
17 |
+
filterbank:
|
18 |
+
kernel_size: 16
|
19 |
+
n_filters: 512
|
20 |
+
stride: 8
|
21 |
+
main_args:
|
22 |
+
exp_dir: exp/wham
|
23 |
+
gpus: -1
|
24 |
+
help: None
|
25 |
+
masknet:
|
26 |
+
bn_chan: 128
|
27 |
+
hid_chan: 512
|
28 |
+
mask_act: relu
|
29 |
+
n_blocks: 8
|
30 |
+
n_repeats: 3
|
31 |
+
n_src: 2
|
32 |
+
skip_chan: 128
|
33 |
+
optim:
|
34 |
+
lr: 0.001
|
35 |
+
optimizer: adam
|
36 |
+
weight_decay: 0.0
|
37 |
+
positional arguments:
|
38 |
+
training:
|
39 |
+
batch_size: 24
|
40 |
+
early_stop: True
|
41 |
+
epochs: 200
|
42 |
+
half_lr: True
|
43 |
+
num_workers: 4
|
44 |
+
```
|
45 |
+
|
46 |
+
### Results:
|
47 |
+
```yaml
|
48 |
+
si_sdr: 16.21326632846293
|
49 |
+
si_sdr_imp: 16.21441705664987
|
50 |
+
sdr: 16.615180021738933
|
51 |
+
sdr_imp: 16.464137807433435
|
52 |
+
sir: 26.860503975131923
|
53 |
+
sir_imp: 26.709461760826414
|
54 |
+
sar: 17.18312813480803
|
55 |
+
sar_imp: -131.99332048277296
|
56 |
+
stoi: 0.9619940905157323
|
57 |
+
stoi_imp: 0.2239480672473015
|
58 |
+
```
|
59 |
+
|
60 |
+
### License notice:
|
61 |
+
This work "ConvTasNet_WHAM!_sepclean" is a derivative of [CSR-I (WSJ0) Complete](https://catalog.ldc.upenn.edu/LDC93S6A)
|
62 |
+
by [LDC](https://www.ldc.upenn.edu/), used under [LDC User Agreement for
|
63 |
+
Non-Members](https://catalog.ldc.upenn.edu/license/ldc-non-members-agreement.pdf) (Research only).
|
64 |
+
"ConvTasNet_WHAM!_sepclean" is licensed under [Attribution-ShareAlike 3.0 Unported](https://creativecommons.org/licenses/by-sa/3.0/)
|
65 |
+
by Manuel Pariente.
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3ad07942ee1a3ae74c47ddc0e215573e813404a1a3eb2625c39e2985a6ca108b
|
3 |
+
size 20272509
|