This is an Audacity wrapper for the model, forked from the repository mpariente/ConvTasNet_WHAM_sepclean,
This model was trained using the Asteroid library:
https://github.com/asteroid-team/asteroid.
This model was trained by Manuel Pariente
using the wham/ConvTasNet recipe in
Asteroid.
It was trained on the
sep_clean task of the WHAM! dataset.
1data:
2 n_src: 2
3 mode: min
4 nondefault_nsrc: None
5 sample_rate: 8000
6 segment: 3
7 task: sep_clean
8 train_dir: data/wav8k/min/tr/
9 valid_dir: data/wav8k/min/cv/
10filterbank:
11 kernel_size: 16
12 n_filters: 512
13 stride: 8
14main_args:
15 exp_dir: exp/wham
16 gpus: -1
17 help: None
18masknet:
19 bn_chan: 128
20 hid_chan: 512
21 mask_act: relu
22 n_blocks: 8
23 n_repeats: 3
24 n_src: 2
25 skip_chan: 128
26optim:
27 lr: 0.001
28 optimizer: adam
29 weight_decay: 0.0
30positional arguments:
31training:
32 batch_size: 24
33 early_stop: True
34 epochs: 200
35 half_lr: True
36 num_workers: 4
1si_sdr: 16.21326632846293
2si_sdr_imp: 16.21441705664987
3sdr: 16.615180021738933
4sdr_imp: 16.464137807433435
5sir: 26.860503975131923
6sir_imp: 26.709461760826414
7sar: 17.18312813480803
8sar_imp: -131.99332048277296
9stoi: 0.9619940905157323
10stoi_imp: 0.2239480672473015
This work "ConvTasNet_WHAM!_sepclean" is a derivative of
CSR-I (WSJ0) Complete
by
LDC, used under
LDC User Agreement for
Non-Members (Research only).
"ConvTasNet_WHAM!_sepclean" is licensed under
Attribution-ShareAlike 3.0 Unported
by Manuel Pariente.
