mpariente/ConvTasNet_WHAM!_sepclean

Description:

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.

Training config:

• data:
  • n_src: 2
  • mode: min
  • nondefault_nsrc: None
  • sample_rate: 8000
  • segment: 3
  • task: sep_clean
  • train_dir: data/wav8k/min/tr/
  • valid_dir: data/wav8k/min/cv/
• filterbank:
  • kernel_size: 16
  • n_filters: 512
  • stride: 8
• main_args:
  • exp_dir: exp/wham
  • gpus: -1
  • help: None
• masknet:
  • bn_chan: 128
  • hid_chan: 512
  • mask_act: relu
  • n_blocks: 8
  • n_repeats: 3
  • n_src: 2
  • skip_chan: 128
• optim:
  • lr: 0.001
  • optimizer: adam
  • weight_decay: 0.0
• positional arguments:
• training:
  • batch_size: 24
  • early_stop: True
  • epochs: 200
  • half_lr: True
  • num_workers: 4

Results:

• si_sdr: 16.21326632846293
• si_sdr_imp: 16.21441705664987
• sdr: 16.615180021738933
• sdr_imp: 16.464137807433435
• sir: 26.860503975131923
• sir_imp: 26.709461760826414
• sar: 17.18312813480803
• sar_imp: -131.99332048277296
• stoi: 0.9619940905157323
• stoi_imp: 0.2239480672473015

License notice:

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.