Siddhant commited on
Commit
60d427a
1 Parent(s): cfdc2b1

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
+ - librimix
10
+ license: cc-by-4.0
11
+ ---
12
+ ## Example ESPnet2 ENH model
13
+ ### `anogkongda/librimix_enh_train_raw_valid.si_snr.ave`
14
+ ♻️ Imported from https://zenodo.org/record/4480771/
15
+
16
+ This model was trained by anogkongda using librimix/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
exp/enh_train_raw/64epoch.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5dc0e50e9ad1089cb8e5fa235e53d26b47e0177a0125abf2a275e6cb3b019f8
3
+ size 13878717
exp/enh_train_raw/RESULTS.TXT ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!-- Generated by ./scripts/utils/show_enh_score.sh -->
2
+ # RESULTS
3
+ ## Environments
4
+ - date: `Mon Jan 25 19:16:45 CST 2021`
5
+ - python version: `3.6.3 |Anaconda, Inc.| (default, Nov 20 2017, 20:41:42) [GCC 7.2.0]`
6
+ - espnet version: `espnet 0.9.7`
7
+ - pytorch version: `pytorch 1.6.0`
8
+ - Git hash: `dcaba2585e28b85c815807165ba9953565ee8694`
9
+ - Commit date: `Thu Jan 21 21:26:59 2021 +0800`
10
+
11
+
12
+ ## enh_train_raw
13
+
14
+ config: ./conf/train.yaml
15
+
16
+ |dataset|STOI|SAR|SDR|SIR|
17
+ |---|---|---|---|---|
18
+ |enhanced_dev|0.845746|11.1029|10.6679|22.6471|
19
+ |enhanced_test|0.846766|10.9166|10.4193|22.0783|
20
+
exp/enh_train_raw/config.yaml ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ config: ./conf/train.yaml
2
+ print_config: false
3
+ log_level: INFO
4
+ dry_run: false
5
+ iterator_type: chunk
6
+ output_dir: exp/enh_train_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: 4
14
+ dist_rank: 0
15
+ local_rank: 0
16
+ dist_master_addr: localhost
17
+ dist_master_port: 48369
18
+ dist_launcher: null
19
+ multiprocessing_distributed: true
20
+ cudnn_enabled: true
21
+ cudnn_benchmark: false
22
+ cudnn_deterministic: true
23
+ collect_stats: false
24
+ write_collected_feats: false
25
+ max_epoch: 200
26
+ patience: 5
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: 16
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
+ - exp/enh_stats_8k/train/noise_ref1_shape
69
+ valid_shape_file:
70
+ - exp/enh_stats_8k/valid/speech_mix_shape
71
+ - exp/enh_stats_8k/valid/speech_ref1_shape
72
+ - exp/enh_stats_8k/valid/speech_ref2_shape
73
+ - exp/enh_stats_8k/valid/noise_ref1_shape
74
+ batch_type: folded
75
+ valid_batch_type: null
76
+ fold_length:
77
+ - 80000
78
+ - 80000
79
+ - 80000
80
+ - 80000
81
+ sort_in_batch: descending
82
+ sort_batch: descending
83
+ multiple_iterator: false
84
+ chunk_length: 24000
85
+ chunk_shift_ratio: 0.5
86
+ num_cache_chunks: 1024
87
+ train_data_path_and_name_and_type:
88
+ - - dump/raw/train/wav.scp
89
+ - speech_mix
90
+ - sound
91
+ - - dump/raw/train/spk1.scp
92
+ - speech_ref1
93
+ - sound
94
+ - - dump/raw/train/spk2.scp
95
+ - speech_ref2
96
+ - sound
97
+ - - dump/raw/train/noise1.scp
98
+ - noise_ref1
99
+ - sound
100
+ valid_data_path_and_name_and_type:
101
+ - - dump/raw/dev/wav.scp
102
+ - speech_mix
103
+ - sound
104
+ - - dump/raw/dev/spk1.scp
105
+ - speech_ref1
106
+ - sound
107
+ - - dump/raw/dev/spk2.scp
108
+ - speech_ref2
109
+ - sound
110
+ - - dump/raw/dev/noise1.scp
111
+ - noise_ref1
112
+ - sound
113
+ allow_variable_data_keys: false
114
+ max_cache_size: 0.0
115
+ max_cache_fd: 32
116
+ valid_max_cache_size: null
117
+ optim: adam
118
+ optim_conf:
119
+ lr: 0.001
120
+ weight_decay: 0
121
+ scheduler: reducelronplateau
122
+ scheduler_conf:
123
+ mode: min
124
+ factor: 0.5
125
+ patience: 1
126
+ init: xavier_uniform
127
+ model_conf:
128
+ loss_type: si_snr
129
+ use_preprocessor: false
130
+ encoder: conv
131
+ encoder_conf:
132
+ channel: 512
133
+ kernel_size: 16
134
+ stride: 8
135
+ separator: tcn
136
+ separator_conf:
137
+ num_spk: 2
138
+ layer: 8
139
+ stack: 3
140
+ bottleneck_dim: 128
141
+ hidden_dim: 512
142
+ kernel: 3
143
+ causal: false
144
+ norm_type: gLN
145
+ nonlinear: relu
146
+ decoder: conv
147
+ decoder_conf:
148
+ channel: 512
149
+ kernel_size: 16
150
+ stride: 8
151
+ required:
152
+ - output_dir
153
+ version: 0.9.7
154
+ distributed: true
exp/enh_train_raw/images/backward_time.png ADDED
Binary file
exp/enh_train_raw/images/forward_time.png ADDED
Binary file
exp/enh_train_raw/images/iter_time.png ADDED
Binary file
exp/enh_train_raw/images/loss.png ADDED
Binary file
exp/enh_train_raw/images/lr_0.png ADDED
Binary file
exp/enh_train_raw/images/optim_step_time.png ADDED
Binary file
exp/enh_train_raw/images/si_snr.png ADDED
Binary file
exp/enh_train_raw/images/train_time.png ADDED
Binary file
meta.yaml ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
1
+ espnet: 0.9.7
2
+ files:
3
+ model_file: exp/enh_train_raw/64epoch.pth
4
+ python: "3.6.3 |Anaconda, Inc.| (default, Nov 20 2017, 20:41:42) \n[GCC 7.2.0]"
5
+ timestamp: 1611988263.681143
6
+ torch: 1.6.0
7
+ yaml_files:
8
+ train_config: exp/enh_train_raw/config.yaml