Update README.md
Browse files
README.md
CHANGED
@@ -16,4 +16,55 @@ It was trained on the vocal ensemble separation task of [the jaCappella dataset]
|
|
16 |
See [the jaCappella dataset page](https://tomohikonakamura.github.io/jaCappella_corpus/).
|
17 |
|
18 |
# Citation
|
19 |
-
See [the jaCappella dataset page](https://tomohikonakamura.github.io/jaCappella_corpus/).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
See [the jaCappella dataset page](https://tomohikonakamura.github.io/jaCappella_corpus/).
|
17 |
|
18 |
# Citation
|
19 |
+
See [the jaCappella dataset page](https://tomohikonakamura.github.io/jaCappella_corpus/).
|
20 |
+
|
21 |
+
# Configuration
|
22 |
+
```yaml
|
23 |
+
data:
|
24 |
+
num_workers: 12
|
25 |
+
sample_rate: 48000
|
26 |
+
samples_per_track: 13
|
27 |
+
seed: 42
|
28 |
+
seq_dur: 5.046
|
29 |
+
source_augmentations:
|
30 |
+
- gain
|
31 |
+
sources:
|
32 |
+
- vocal_percussion
|
33 |
+
- bass
|
34 |
+
- alto
|
35 |
+
- tenor
|
36 |
+
- soprano
|
37 |
+
- lead_vocal
|
38 |
+
filterbank:
|
39 |
+
kernel_size: 32
|
40 |
+
n_filters: 64
|
41 |
+
stride: 16
|
42 |
+
masknet:
|
43 |
+
bidirectional: true
|
44 |
+
chunk_size: 174
|
45 |
+
dropout: 0
|
46 |
+
ff_activation: relu
|
47 |
+
ff_hid: 256
|
48 |
+
hop_size: 128
|
49 |
+
in_chan: 64
|
50 |
+
mask_act: sigmoid
|
51 |
+
n_repeats: 8
|
52 |
+
n_src: 6
|
53 |
+
norm_type: gLN
|
54 |
+
out_chan: 64
|
55 |
+
optim:
|
56 |
+
lr: 0.005
|
57 |
+
optimizer: adam
|
58 |
+
weight_decay: 1.0e-05
|
59 |
+
training:
|
60 |
+
batch_size: 1
|
61 |
+
early_stop: true
|
62 |
+
epochs: 600
|
63 |
+
gradient_clipping: 5
|
64 |
+
half_lr: true
|
65 |
+
loss_func: pit_sisdr
|
66 |
+
```
|
67 |
+
# Results on vocal ensemble separation
|
68 |
+
| Method | Lead vocal | Soprano | Alto | Tenor | Bass |Vocal percussion|
|
69 |
+
|:---------------:|:--------------:|:--------------:|:--------------:|:--------------:|:--------------:|:--------------:|
|
70 |
+
| DPTNet | 8.9 | 8.5 | 11.9 | 14.9 | 19.7 | 21.9 |
|