metadata
tags:
- asteroid
- audio
- ConvTasNet
- audio-to-audio
datasets:
- Libri2Mix
- sep_noisy
license: cc-by-sa-4.0
Training config:
data:
n_src: 2
sample_rate: 8000
segment: 3
task: sep_noisy
train_dir: data/wav8k/min/train-360
valid_dir: data/wav8k/min/dev
filterbank:
kernel_size: 16
n_filters: 512
stride: 8
masknet:
bn_chan: 128
hid_chan: 512
mask_act: relu
n_blocks: 8
n_repeats: 3
skip_chan: 128
optim:
lr: 0.001
optimizer: adam
weight_decay: 0.0
training:
batch_size: 24
early_stop: True
epochs: 200
half_lr: True
num_workers: 4
Results:
On Libri2Mix min test set :
si_sdr: 9.944424856077259
si_sdr_imp: 11.939395359731192
sdr: 10.701526190782072
sdr_imp: 12.481757547845662
sir: 22.633644975545575
sir_imp: 22.45666740833025
sar: 11.131644100944868
sar_imp: 4.248489589311784
stoi: 0.852048619949357
stoi_imp: 0.2071994899565506