Update README.md
Browse files
README.md
CHANGED
@@ -6,23 +6,64 @@ sample_rate: 8000
|
|
6 |
|
7 |
---
|
8 |
|
9 |
-
This is an Audacity wrapper for the model, forked from the repository groadabike/ConvTasNet_DAMP-VSEP_enhboth
|
10 |
This model was trained using the Asteroid library: https://github.com/asteroid-team/asteroid.
|
11 |
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
---
|
8 |
|
9 |
+
This is an Audacity wrapper for the model, forked from the repository `groadabike/ConvTasNet_DAMP-VSEP_enhboth`,
|
10 |
This model was trained using the Asteroid library: https://github.com/asteroid-team/asteroid.
|
11 |
|
12 |
+
The following info was copied directly from `groadabike/ConvTasNet_DAMP-VSEP_enhboth`:
|
13 |
+
|
14 |
+
### Description:
|
15 |
+
This model was trained by Gerardo Roa Dabike using Asteroid. It was trained on the enh_both task of the DAMP-VSEP dataset.
|
16 |
+
### Training config:
|
17 |
+
```yaml
|
18 |
+
data:
|
19 |
+
channels: 1
|
20 |
+
n_src: 2
|
21 |
+
root_path: data
|
22 |
+
sample_rate: 16000
|
23 |
+
samples_per_track: 10
|
24 |
+
segment: 3.0
|
25 |
+
task: enh_both
|
26 |
+
filterbank:
|
27 |
+
kernel_size: 20
|
28 |
+
n_filters: 256
|
29 |
+
stride: 10
|
30 |
+
main_args:
|
31 |
+
exp_dir: exp/train_convtasnet
|
32 |
+
help: None
|
33 |
+
masknet:
|
34 |
+
bn_chan: 256
|
35 |
+
conv_kernel_size: 3
|
36 |
+
hid_chan: 512
|
37 |
+
mask_act: relu
|
38 |
+
n_blocks: 8
|
39 |
+
n_repeats: 4
|
40 |
+
n_src: 2
|
41 |
+
norm_type: gLN
|
42 |
+
skip_chan: 256
|
43 |
+
optim:
|
44 |
+
lr: 0.0003
|
45 |
+
optimizer: adam
|
46 |
+
weight_decay: 0.0
|
47 |
+
positional arguments:
|
48 |
+
training:
|
49 |
+
batch_size: 12
|
50 |
+
early_stop: True
|
51 |
+
epochs: 50
|
52 |
+
half_lr: True
|
53 |
+
num_workers: 12
|
54 |
+
```
|
55 |
+
### Results:
|
56 |
+
```yaml
|
57 |
+
si_sdr: 14.018196157142519
|
58 |
+
si_sdr_imp: 14.017103133809577
|
59 |
+
sdr: 14.498517291333885
|
60 |
+
sdr_imp: 14.463389151567865
|
61 |
+
sir: 24.149634529133372
|
62 |
+
sir_imp: 24.11450638936735
|
63 |
+
sar: 15.338597389045935
|
64 |
+
sar_imp: -137.30634122401517
|
65 |
+
stoi: 0.7639416744417206
|
66 |
+
stoi_imp: 0.1843383526963759
|
67 |
+
```
|
68 |
+
### License notice:
|
69 |
+
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.
|