Samuele Cornell/FasNetTAC_TACDataset_separatenoisy

Description:

This model was trained by popcornell using the TAC/TAC recipe in Asteroid. It was trained on the separate_noisy task of the TACDataset dataset.

 

Training config:

• data:
  • dev_json: ./data/validation.json
  • sample_rate: 16000
  • segment: None
  • test_json: ./data/test.json
  • train_json: ./data/train.json
• main_args:
  • exp_dir: exp/train_TAC_publish
  • help: None
• net:
  • chunk_size: 50
  • context_ms: 16
  • enc_dim: 64
  • feature_dim: 64
  • hidden_dim: 128
  • hop_size: 25
  • n_layers: 4
  • n_src: 2
  • window_ms: 4
• optim:
  • lr: 0.001
  • weight_decay: 1e-06
• positional arguments:
• training:
  • accumulate_batches: 1
  • batch_size: 8
  • early_stop: True
  • epochs: 200
  • gradient_clipping: 5
  • half_lr: True
  • num_workers: 8
  • patience: 30
  • save_top_k: 10

 

Results:

• si_sdr: 10.871864315894744
• si_sdr_imp: 11.322284052560262

 

Versions:

• torch_version: 1.7.1
• pytorch_lightning_version: 1.1.6
• asteroid_version: 0.4.1

 

License notice:

This work "FasNetTAC_TACDataset_separatenoisy" is a derivative of LibriSpeech ASR corpus by Vassil Panayotov, used under CC BY 4.0; of End-to-end Microphone Permutation and Number Invariant Multi-channel Speech Separation by Yi Luo, Zhuo Chen, Nima Mesgarani, Takuya Yoshioka, used under CC BY 4.0. "FasNetTAC_TACDataset_separatenoisy" is licensed under Attribution-ShareAlike 3.0 Unported by popcornell.