Siddhant commited on
Commit
09907bb
1 Parent(s): 0613b2b

import from zenodo

Browse files
README.md ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - espnet
4
+ - audio
5
+ - speech-enhancement
6
+ - audio-to-audio
7
+ language: en
8
+ datasets:
9
+ - wsj0_2mix
10
+ license: cc-by-4.0
11
+ ---
12
+ ## Example ESPnet2 ENH model
13
+ ### `Chenda_Li/wsj0_2mix_enh_train_enh_conv_tasnet_raw_valid.si_snr.ave`
14
+ ♻️ Imported from https://zenodo.org/record/4498562/
15
+
16
+ This model was trained by Chenda Li using wsj0_2mix/enh1 recipe in [espnet](https://github.com/espnet/espnet/).
17
+ ### Demo: How to use in ESPnet2
18
+ ```python
19
+ # coming soon
20
+ ```
21
+ ### Citing ESPnet
22
+ ```BibTex
23
+ @inproceedings{watanabe2018espnet,
24
+ author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson {Enrique Yalta Soplin} and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
25
+ title={{ESPnet}: End-to-End Speech Processing Toolkit},
26
+ year={2018},
27
+ booktitle={Proceedings of Interspeech},
28
+ pages={2207--2211},
29
+ doi={10.21437/Interspeech.2018-1456},
30
+ url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
31
+ }
32
+ @inproceedings{hayashi2020espnet,
33
+ title={{Espnet-TTS}: Unified, reproducible, and integratable open source end-to-end text-to-speech toolkit},
34
+ author={Hayashi, Tomoki and Yamamoto, Ryuichi and Inoue, Katsuki and Yoshimura, Takenori and Watanabe, Shinji and Toda, Tomoki and Takeda, Kazuya and Zhang, Yu and Tan, Xu},
35
+ booktitle={Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
36
+ pages={7654--7658},
37
+ year={2020},
38
+ organization={IEEE}
39
+ }
40
+ ```
41
+ or arXiv:
42
+ ```bibtex
43
+ @misc{watanabe2018espnet,
44
+ title={ESPnet: End-to-End Speech Processing Toolkit},
45
+ author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Enrique Yalta Soplin and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
46
+ year={2018},
47
+ eprint={1804.00015},
48
+ archivePrefix={arXiv},
49
+ primaryClass={cs.CL}
50
+ }
51
+ ```
exp/enh_stats_8k/train/feats_stats.npz ADDED
Binary file (778 Bytes). View file
exp/enh_train_enh_conv_tasnet_raw/66epoch.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a324193d6a4537be5711e00caf18e631df5e10c88896a388e5745626bfe22ae
3
+ size 34969477
exp/enh_train_enh_conv_tasnet_raw/RESULTS.md ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!-- Generated by ./scripts/utils/show_enh_score.sh -->
2
+ # RESULTS
3
+ ## Environments
4
+ - date: `Thu Feb 4 01:16:18 CST 2021`
5
+ - python version: `3.7.6 (default, Jan 8 2020, 19:59:22) [GCC 7.3.0]`
6
+ - espnet version: `espnet 0.9.7`
7
+ - pytorch version: `pytorch 1.5.0`
8
+ - Git hash: `a3334220b0352931677946d178fade3313cf82bb`
9
+ - Commit date: `Fri Jan 29 23:35:47 2021 +0800`
10
+
11
+
12
+ ## enh_train_enh_conv_tasnet_raw
13
+
14
+ config: ./conf/tuning/train_enh_conv_tasnet.yaml
15
+
16
+ |dataset|STOI|SAR|SDR|SIR|
17
+ |---|---|---|---|---|
18
+ |enhanced_cv_min_8k|0.949205|17.3785|16.8028|26.9785|
19
+ |enhanced_tt_min_8k|0.95349|16.6221|15.9494|25.9032|
20
+
exp/enh_train_enh_conv_tasnet_raw/config.yaml ADDED
@@ -0,0 +1,146 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ config: ./conf/tuning/train_enh_conv_tasnet.yaml
2
+ print_config: false
3
+ log_level: INFO
4
+ dry_run: false
5
+ iterator_type: chunk
6
+ output_dir: exp/enh_train_enh_conv_tasnet_raw
7
+ ngpu: 1
8
+ seed: 0
9
+ num_workers: 4
10
+ num_att_plot: 3
11
+ dist_backend: nccl
12
+ dist_init_method: env://
13
+ dist_world_size: null
14
+ dist_rank: null
15
+ local_rank: 0
16
+ dist_master_addr: null
17
+ dist_master_port: null
18
+ dist_launcher: null
19
+ multiprocessing_distributed: false
20
+ cudnn_enabled: true
21
+ cudnn_benchmark: false
22
+ cudnn_deterministic: true
23
+ collect_stats: false
24
+ write_collected_feats: false
25
+ max_epoch: 100
26
+ patience: 4
27
+ val_scheduler_criterion:
28
+ - valid
29
+ - loss
30
+ early_stopping_criterion:
31
+ - valid
32
+ - loss
33
+ - min
34
+ best_model_criterion:
35
+ - - valid
36
+ - si_snr
37
+ - max
38
+ - - valid
39
+ - loss
40
+ - min
41
+ keep_nbest_models: 1
42
+ grad_clip: 5.0
43
+ grad_clip_type: 2.0
44
+ grad_noise: false
45
+ accum_grad: 1
46
+ no_forward_run: false
47
+ resume: true
48
+ train_dtype: float32
49
+ use_amp: false
50
+ log_interval: null
51
+ unused_parameters: false
52
+ use_tensorboard: true
53
+ use_wandb: false
54
+ wandb_project: null
55
+ wandb_id: null
56
+ pretrain_path: null
57
+ init_param: []
58
+ freeze_param: []
59
+ num_iters_per_epoch: null
60
+ batch_size: 8
61
+ valid_batch_size: null
62
+ batch_bins: 1000000
63
+ valid_batch_bins: null
64
+ train_shape_file:
65
+ - exp/enh_stats_8k/train/speech_mix_shape
66
+ - exp/enh_stats_8k/train/speech_ref1_shape
67
+ - exp/enh_stats_8k/train/speech_ref2_shape
68
+ valid_shape_file:
69
+ - exp/enh_stats_8k/valid/speech_mix_shape
70
+ - exp/enh_stats_8k/valid/speech_ref1_shape
71
+ - exp/enh_stats_8k/valid/speech_ref2_shape
72
+ batch_type: folded
73
+ valid_batch_type: null
74
+ fold_length:
75
+ - 80000
76
+ - 80000
77
+ - 80000
78
+ sort_in_batch: descending
79
+ sort_batch: descending
80
+ multiple_iterator: false
81
+ chunk_length: 32000
82
+ chunk_shift_ratio: 0.5
83
+ num_cache_chunks: 1024
84
+ train_data_path_and_name_and_type:
85
+ - - dump/raw/tr_min_8k/wav.scp
86
+ - speech_mix
87
+ - sound
88
+ - - dump/raw/tr_min_8k/spk1.scp
89
+ - speech_ref1
90
+ - sound
91
+ - - dump/raw/tr_min_8k/spk2.scp
92
+ - speech_ref2
93
+ - sound
94
+ valid_data_path_and_name_and_type:
95
+ - - dump/raw/cv_min_8k/wav.scp
96
+ - speech_mix
97
+ - sound
98
+ - - dump/raw/cv_min_8k/spk1.scp
99
+ - speech_ref1
100
+ - sound
101
+ - - dump/raw/cv_min_8k/spk2.scp
102
+ - speech_ref2
103
+ - sound
104
+ allow_variable_data_keys: false
105
+ max_cache_size: 0.0
106
+ max_cache_fd: 32
107
+ valid_max_cache_size: null
108
+ optim: adam
109
+ optim_conf:
110
+ lr: 0.001
111
+ eps: 1.0e-08
112
+ weight_decay: 0
113
+ scheduler: reducelronplateau
114
+ scheduler_conf:
115
+ mode: min
116
+ factor: 0.5
117
+ patience: 1
118
+ init: xavier_uniform
119
+ model_conf:
120
+ loss_type: si_snr
121
+ use_preprocessor: false
122
+ encoder: conv
123
+ encoder_conf:
124
+ channel: 256
125
+ kernel_size: 20
126
+ stride: 10
127
+ separator: tcn
128
+ separator_conf:
129
+ num_spk: 2
130
+ layer: 8
131
+ stack: 4
132
+ bottleneck_dim: 256
133
+ hidden_dim: 512
134
+ kernel: 3
135
+ causal: false
136
+ norm_type: gLN
137
+ nonlinear: relu
138
+ decoder: conv
139
+ decoder_conf:
140
+ channel: 256
141
+ kernel_size: 20
142
+ stride: 10
143
+ required:
144
+ - output_dir
145
+ version: 0.9.7
146
+ distributed: false
exp/enh_train_enh_conv_tasnet_raw/images/backward_time.png ADDED
Binary file (37.3 kB). View file
exp/enh_train_enh_conv_tasnet_raw/images/forward_time.png ADDED
Binary file (36.4 kB). View file
exp/enh_train_enh_conv_tasnet_raw/images/iter_time.png ADDED
Binary file (32.7 kB). View file
exp/enh_train_enh_conv_tasnet_raw/images/loss.png ADDED
Binary file (33.7 kB). View file
exp/enh_train_enh_conv_tasnet_raw/images/lr_0.png ADDED
Binary file (24.3 kB). View file
exp/enh_train_enh_conv_tasnet_raw/images/optim_step_time.png ADDED
Binary file (38.7 kB). View file
exp/enh_train_enh_conv_tasnet_raw/images/si_snr.png ADDED
Binary file (32.6 kB). View file
exp/enh_train_enh_conv_tasnet_raw/images/train_time.png ADDED
Binary file (28.8 kB). View file
meta.yaml ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
1
+ espnet: 0.9.7
2
+ files:
3
+ model_file: exp/enh_train_enh_conv_tasnet_raw/66epoch.pth
4
+ python: "3.7.6 (default, Jan 8 2020, 19:59:22) \n[GCC 7.3.0]"
5
+ timestamp: 1612372579.804363
6
+ torch: 1.5.0
7
+ yaml_files:
8
+ train_config: exp/enh_train_enh_conv_tasnet_raw/config.yaml