Initial commit
Browse files- README.md +84 -0
- pytorch_model.bin +3 -0
README.md
ADDED
|
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- asteroid
|
| 4 |
+
- audio
|
| 5 |
+
- DPRNNTasNet
|
| 6 |
+
datasets:
|
| 7 |
+
- wham
|
| 8 |
+
- sep_clean
|
| 9 |
+
license: cc-by-sa-3.0
|
| 10 |
+
inference: false
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
## Asteroid model `mpariente/DPRNNTasNet-ks2_WHAM_sepclean`
|
| 14 |
+
Imported from [Zenodo](https://zenodo.org/record/3862942)
|
| 15 |
+
|
| 16 |
+
### Description:
|
| 17 |
+
This model was trained by Manuel Pariente
|
| 18 |
+
using the wham/DPRNN recipe in [Asteroid](https://github.com/asteroid-team/asteroid).
|
| 19 |
+
It was trained on the `sep_clean` task of the WHAM! dataset.
|
| 20 |
+
|
| 21 |
+
### Training config:
|
| 22 |
+
```yaml
|
| 23 |
+
data:
|
| 24 |
+
mode: min
|
| 25 |
+
nondefault_nsrc: None
|
| 26 |
+
sample_rate: 8000
|
| 27 |
+
segment: 2.0
|
| 28 |
+
task: sep_clean
|
| 29 |
+
train_dir: data/wav8k/min/tr
|
| 30 |
+
valid_dir: data/wav8k/min/cv
|
| 31 |
+
filterbank:
|
| 32 |
+
kernel_size: 2
|
| 33 |
+
n_filters: 64
|
| 34 |
+
stride: 1
|
| 35 |
+
main_args:
|
| 36 |
+
exp_dir: exp/train_dprnn_new/
|
| 37 |
+
gpus: -1
|
| 38 |
+
help: None
|
| 39 |
+
masknet:
|
| 40 |
+
bidirectional: True
|
| 41 |
+
bn_chan: 128
|
| 42 |
+
chunk_size: 250
|
| 43 |
+
dropout: 0
|
| 44 |
+
hid_size: 128
|
| 45 |
+
hop_size: 125
|
| 46 |
+
in_chan: 64
|
| 47 |
+
mask_act: sigmoid
|
| 48 |
+
n_repeats: 6
|
| 49 |
+
n_src: 2
|
| 50 |
+
out_chan: 64
|
| 51 |
+
optim:
|
| 52 |
+
lr: 0.001
|
| 53 |
+
optimizer: adam
|
| 54 |
+
weight_decay: 1e-05
|
| 55 |
+
positional arguments:
|
| 56 |
+
training:
|
| 57 |
+
batch_size: 3
|
| 58 |
+
early_stop: True
|
| 59 |
+
epochs: 200
|
| 60 |
+
gradient_clipping: 5
|
| 61 |
+
half_lr: True
|
| 62 |
+
num_workers: 8
|
| 63 |
+
```
|
| 64 |
+
|
| 65 |
+
### Results:
|
| 66 |
+
```yaml
|
| 67 |
+
si_sdr: 19.316743490695334
|
| 68 |
+
si_sdr_imp: 19.317895273889842
|
| 69 |
+
sdr: 19.68085347190952
|
| 70 |
+
sdr_imp: 19.5298092932871
|
| 71 |
+
sir: 30.362213998701232
|
| 72 |
+
sir_imp: 30.21116982007881
|
| 73 |
+
sar: 20.15553251343315
|
| 74 |
+
sar_imp: -129.02091762351188
|
| 75 |
+
stoi: 0.97772664309074
|
| 76 |
+
stoi_imp: 0.23968091518217424
|
| 77 |
+
```
|
| 78 |
+
|
| 79 |
+
### License notice:
|
| 80 |
+
This work "DPRNNTasNet-ks2_WHAM_sepclean" is a derivative of [CSR-I (WSJ0) Complete](https://catalog.ldc.upenn.edu/LDC93S6A)
|
| 81 |
+
by [LDC](https://www.ldc.upenn.edu/), used under [LDC User Agreement for
|
| 82 |
+
Non-Members](https://catalog.ldc.upenn.edu/license/ldc-non-members-agreement.pdf) (Research only).
|
| 83 |
+
"DPRNNTasNet-ks2_WHAM_sepclean" is licensed under [Attribution-ShareAlike 3.0 Unported](https://creativecommons.org/licenses/by-sa/3.0/)
|
| 84 |
+
by Manuel Pariente.
|
pytorch_model.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cc1f1fecf24ea3e486521029dc0e1444686bd4b6fdf9715e7757936cbd9ffdf6
|
| 3 |
+
size 14664381
|