Update README.md
Browse files
README.md
CHANGED
|
@@ -4,4 +4,69 @@ tags:
|
|
| 4 |
inference: false
|
| 5 |
---
|
| 6 |
This is an Audacity wrapper for the model, forked from the repository mpariente/ConvTasNet_WHAM_sepclean,
|
| 7 |
-
This model was trained using the Asteroid library: https://github.com/asteroid-team/asteroid.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
inference: false
|
| 5 |
---
|
| 6 |
This is an Audacity wrapper for the model, forked from the repository mpariente/ConvTasNet_WHAM_sepclean,
|
| 7 |
+
This model was trained using the Asteroid library: https://github.com/asteroid-team/asteroid.
|
| 8 |
+
|
| 9 |
+
The following info was copied from `mpariente/ConvTasNet_WHAM_sepclean`:
|
| 10 |
+
|
| 11 |
+
### Description:
|
| 12 |
+
This model was trained by Manuel Pariente
|
| 13 |
+
using the wham/ConvTasNet recipe in [Asteroid](https://github.com/asteroid-team/asteroid).
|
| 14 |
+
It was trained on the `sep_clean` task of the WHAM! dataset.
|
| 15 |
+
### Training config:
|
| 16 |
+
```yaml
|
| 17 |
+
data:
|
| 18 |
+
n_src: 2
|
| 19 |
+
mode: min
|
| 20 |
+
nondefault_nsrc: None
|
| 21 |
+
sample_rate: 8000
|
| 22 |
+
segment: 3
|
| 23 |
+
task: sep_clean
|
| 24 |
+
train_dir: data/wav8k/min/tr/
|
| 25 |
+
valid_dir: data/wav8k/min/cv/
|
| 26 |
+
filterbank:
|
| 27 |
+
kernel_size: 16
|
| 28 |
+
n_filters: 512
|
| 29 |
+
stride: 8
|
| 30 |
+
main_args:
|
| 31 |
+
exp_dir: exp/wham
|
| 32 |
+
gpus: -1
|
| 33 |
+
help: None
|
| 34 |
+
masknet:
|
| 35 |
+
bn_chan: 128
|
| 36 |
+
hid_chan: 512
|
| 37 |
+
mask_act: relu
|
| 38 |
+
n_blocks: 8
|
| 39 |
+
n_repeats: 3
|
| 40 |
+
n_src: 2
|
| 41 |
+
skip_chan: 128
|
| 42 |
+
optim:
|
| 43 |
+
lr: 0.001
|
| 44 |
+
optimizer: adam
|
| 45 |
+
weight_decay: 0.0
|
| 46 |
+
positional arguments:
|
| 47 |
+
training:
|
| 48 |
+
batch_size: 24
|
| 49 |
+
early_stop: True
|
| 50 |
+
epochs: 200
|
| 51 |
+
half_lr: True
|
| 52 |
+
num_workers: 4
|
| 53 |
+
```
|
| 54 |
+
### Results:
|
| 55 |
+
```yaml
|
| 56 |
+
si_sdr: 16.21326632846293
|
| 57 |
+
si_sdr_imp: 16.21441705664987
|
| 58 |
+
sdr: 16.615180021738933
|
| 59 |
+
sdr_imp: 16.464137807433435
|
| 60 |
+
sir: 26.860503975131923
|
| 61 |
+
sir_imp: 26.709461760826414
|
| 62 |
+
sar: 17.18312813480803
|
| 63 |
+
sar_imp: -131.99332048277296
|
| 64 |
+
stoi: 0.9619940905157323
|
| 65 |
+
stoi_imp: 0.2239480672473015
|
| 66 |
+
```
|
| 67 |
+
### License notice:
|
| 68 |
+
This work "ConvTasNet_WHAM!_sepclean" is a derivative of [CSR-I (WSJ0) Complete](https://catalog.ldc.upenn.edu/LDC93S6A)
|
| 69 |
+
by [LDC](https://www.ldc.upenn.edu/), used under [LDC User Agreement for
|
| 70 |
+
Non-Members](https://catalog.ldc.upenn.edu/license/ldc-non-members-agreement.pdf) (Research only).
|
| 71 |
+
"ConvTasNet_WHAM!_sepclean" is licensed under [Attribution-ShareAlike 3.0 Unported](https://creativecommons.org/licenses/by-sa/3.0/)
|
| 72 |
+
by Manuel Pariente.
|