Julien Chaumond
commited on
Commit
·
efc58bf
1
Parent(s):
7255565
initial import from https://zenodo.org/record/3903795#.X8pMBRNKjUI
Browse files- README.md +78 -0
- pytorch_model.bin +3 -0
README.md
ADDED
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
tags:
|
3 |
+
- asteroid
|
4 |
+
- audio
|
5 |
+
datasets:
|
6 |
+
- wham
|
7 |
+
- sep_clean
|
8 |
+
inference: false
|
9 |
+
---
|
10 |
+
|
11 |
+
## Asteroid model `mpariente/DPRNNTasNet(ks=16)_WHAM!_sepclean`
|
12 |
+
|
13 |
+
⚠️ Imported from https://zenodo.org/record/3903795#.X8pMBRNKjUI
|
14 |
+
|
15 |
+
This model was trained by Manuel Pariente using the wham/DPRNN recipe in Asteroid. It was trained on the sep_clean task of the WHAM! dataset.
|
16 |
+
|
17 |
+
|
18 |
+
### Demo: How to use in Asteroid
|
19 |
+
|
20 |
+
```python
|
21 |
+
# coming soon
|
22 |
+
```
|
23 |
+
|
24 |
+
|
25 |
+
### Training config
|
26 |
+
|
27 |
+
- data:
|
28 |
+
- mode: min
|
29 |
+
- nondefault_nsrc: None
|
30 |
+
- sample_rate: 8000
|
31 |
+
- segment: 2.0
|
32 |
+
- task: sep_clean
|
33 |
+
- train_dir: data/wav8k/min/tr
|
34 |
+
- valid_dir: data/wav8k/min/cv
|
35 |
+
- filterbank:
|
36 |
+
- kernel_size: 16
|
37 |
+
- n_filters: 64
|
38 |
+
- stride: 8
|
39 |
+
- main_args:
|
40 |
+
- exp_dir: exp/train_dprnn_ks16/
|
41 |
+
- help: None
|
42 |
+
- masknet:
|
43 |
+
- bidirectional: True
|
44 |
+
- bn_chan: 128
|
45 |
+
- chunk_size: 100
|
46 |
+
- dropout: 0
|
47 |
+
- hid_size: 128
|
48 |
+
- hop_size: 50
|
49 |
+
- in_chan: 64
|
50 |
+
- mask_act: sigmoid
|
51 |
+
- n_repeats: 6
|
52 |
+
- n_src: 2
|
53 |
+
- out_chan: 64
|
54 |
+
- optim:
|
55 |
+
- lr: 0.001
|
56 |
+
- optimizer: adam
|
57 |
+
- weight_decay: 1e-05
|
58 |
+
- positional arguments:
|
59 |
+
- training:
|
60 |
+
- batch_size: 6
|
61 |
+
- early_stop: True
|
62 |
+
- epochs: 200
|
63 |
+
- gradient_clipping: 5
|
64 |
+
- half_lr: True
|
65 |
+
- num_workers: 6
|
66 |
+
|
67 |
+
#### Results
|
68 |
+
|
69 |
+
- `si_sdr`: 18.227683982688003
|
70 |
+
- `si_sdr_imp`: 18.22883576588251
|
71 |
+
- `sdr`: 18.617789605060587
|
72 |
+
- `sdr_imp`: 18.466745426438173
|
73 |
+
- `sir`: 29.22773720052717
|
74 |
+
- `sir_imp`: 29.07669302190474
|
75 |
+
- `sar`: 19.116352171914485
|
76 |
+
- `sar_imp`: -130.06009796503054
|
77 |
+
- `stoi`: 0.9722025377865715
|
78 |
+
- `stoi_imp`: 0.23415680987800583
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:97cab56c48c6b86a2bcbd32a8b7ca8ac9b9fc5b2621c991f26f633d7cf14c182
|
3 |
+
size 14644200
|