Siddhant commited on
Commit
1328b49
1 Parent(s): d95ce92

import from zenodo

Browse files
README.md ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - espnet
4
+ - audio
5
+ - text-to-speech
6
+ language: ja
7
+ datasets:
8
+ - jsut
9
+ license: cc-by-4.0
10
+ ---
11
+ ## Example ESPnet2 TTS model
12
+ ### `kan-bayashi/jsut_transformer`
13
+ ♻️ Imported from https://zenodo.org/record/4034121/
14
+
15
+ This model was trained by kan-bayashi using jsut/tts1 recipe in [espnet](https://github.com/espnet/espnet/).
16
+ ### Demo: How to use in ESPnet2
17
+ ```python
18
+ # coming soon
19
+ ```
20
+ ### Citing ESPnet
21
+ ```BibTex
22
+ @inproceedings{watanabe2018espnet,
23
+ 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},
24
+ title={{ESPnet}: End-to-End Speech Processing Toolkit},
25
+ year={2018},
26
+ booktitle={Proceedings of Interspeech},
27
+ pages={2207--2211},
28
+ doi={10.21437/Interspeech.2018-1456},
29
+ url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
30
+ }
31
+ @inproceedings{hayashi2020espnet,
32
+ title={{Espnet-TTS}: Unified, reproducible, and integratable open source end-to-end text-to-speech toolkit},
33
+ 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},
34
+ booktitle={Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
35
+ pages={7654--7658},
36
+ year={2020},
37
+ organization={IEEE}
38
+ }
39
+ ```
40
+ or arXiv:
41
+ ```bibtex
42
+ @misc{watanabe2018espnet,
43
+ title={ESPnet: End-to-End Speech Processing Toolkit},
44
+ 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},
45
+ year={2018},
46
+ eprint={1804.00015},
47
+ archivePrefix={arXiv},
48
+ primaryClass={cs.CL}
49
+ }
50
+ ```
exp/tts_stats_raw_phn_jaconv_pyopenjtalk/train/feats_stats.npz ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/config.yaml ADDED
@@ -0,0 +1,218 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ config: conf/tuning/train_transformer.yaml
2
+ print_config: false
3
+ log_level: INFO
4
+ dry_run: false
5
+ iterator_type: sequence
6
+ output_dir: exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk
7
+ ngpu: 1
8
+ seed: 0
9
+ num_workers: 1
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: 58206
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: null
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
+ - loss
37
+ - min
38
+ - - train
39
+ - loss
40
+ - min
41
+ keep_nbest_models: 5
42
+ grad_clip: 1.0
43
+ grad_clip_type: 2.0
44
+ grad_noise: false
45
+ accum_grad: 2
46
+ no_forward_run: false
47
+ resume: true
48
+ train_dtype: float32
49
+ use_amp: false
50
+ log_interval: null
51
+ pretrain_path: []
52
+ pretrain_key: []
53
+ num_iters_per_epoch: 1000
54
+ batch_size: 20
55
+ valid_batch_size: null
56
+ batch_bins: 9000000
57
+ valid_batch_bins: null
58
+ train_shape_file:
59
+ - exp/tts_stats_raw_phn_jaconv_pyopenjtalk/train/text_shape.phn
60
+ - exp/tts_stats_raw_phn_jaconv_pyopenjtalk/train/speech_shape
61
+ valid_shape_file:
62
+ - exp/tts_stats_raw_phn_jaconv_pyopenjtalk/valid/text_shape.phn
63
+ - exp/tts_stats_raw_phn_jaconv_pyopenjtalk/valid/speech_shape
64
+ batch_type: numel
65
+ valid_batch_type: null
66
+ fold_length:
67
+ - 150
68
+ - 240000
69
+ sort_in_batch: descending
70
+ sort_batch: descending
71
+ multiple_iterator: false
72
+ chunk_length: 500
73
+ chunk_shift_ratio: 0.5
74
+ num_cache_chunks: 1024
75
+ train_data_path_and_name_and_type:
76
+ - - dump/raw/tr_no_dev/text
77
+ - text
78
+ - text
79
+ - - dump/raw/tr_no_dev/wav.scp
80
+ - speech
81
+ - sound
82
+ valid_data_path_and_name_and_type:
83
+ - - dump/raw/dev/text
84
+ - text
85
+ - text
86
+ - - dump/raw/dev/wav.scp
87
+ - speech
88
+ - sound
89
+ allow_variable_data_keys: false
90
+ max_cache_size: 0.0
91
+ valid_max_cache_size: null
92
+ optim: adam
93
+ optim_conf:
94
+ lr: 1.0
95
+ scheduler: noamlr
96
+ scheduler_conf:
97
+ model_size: 512
98
+ warmup_steps: 8000
99
+ token_list:
100
+ - <blank>
101
+ - <unk>
102
+ - a
103
+ - o
104
+ - i
105
+ - u
106
+ - e
107
+ - k
108
+ - n
109
+ - t
110
+ - r
111
+ - s
112
+ - N
113
+ - m
114
+ - pau
115
+ - sh
116
+ - d
117
+ - g
118
+ - w
119
+ - U
120
+ - I
121
+ - cl
122
+ - h
123
+ - y
124
+ - b
125
+ - j
126
+ - ts
127
+ - ch
128
+ - z
129
+ - p
130
+ - f
131
+ - ky
132
+ - ry
133
+ - gy
134
+ - hy
135
+ - ny
136
+ - by
137
+ - my
138
+ - py
139
+ - v
140
+ - dy
141
+ - ty
142
+ - <sos/eos>
143
+ odim: null
144
+ model_conf: {}
145
+ use_preprocessor: true
146
+ token_type: phn
147
+ bpemodel: null
148
+ non_linguistic_symbols: null
149
+ cleaner: jaconv
150
+ g2p: pyopenjtalk
151
+ feats_extract: fbank
152
+ feats_extract_conf:
153
+ fs: 24000
154
+ fmin: 80
155
+ fmax: 7600
156
+ n_mels: 80
157
+ hop_length: 300
158
+ n_fft: 2048
159
+ win_length: 1200
160
+ normalize: global_mvn
161
+ normalize_conf:
162
+ stats_file: exp/tts_stats_raw_phn_jaconv_pyopenjtalk/train/feats_stats.npz
163
+ tts: transformer
164
+ tts_conf:
165
+ embed_dim: 0
166
+ eprenet_conv_layers: 0
167
+ eprenet_conv_filts: 0
168
+ eprenet_conv_chans: 0
169
+ dprenet_layers: 2
170
+ dprenet_units: 256
171
+ adim: 512
172
+ aheads: 8
173
+ elayers: 6
174
+ eunits: 1024
175
+ dlayers: 6
176
+ dunits: 1024
177
+ positionwise_layer_type: conv1d
178
+ positionwise_conv_kernel_size: 1
179
+ postnet_layers: 5
180
+ postnet_filts: 5
181
+ postnet_chans: 256
182
+ use_masking: true
183
+ bce_pos_weight: 5.0
184
+ use_scaled_pos_enc: true
185
+ encoder_normalize_before: false
186
+ decoder_normalize_before: false
187
+ reduction_factor: 1
188
+ init_type: xavier_uniform
189
+ init_enc_alpha: 1.0
190
+ init_dec_alpha: 1.0
191
+ eprenet_dropout_rate: 0.0
192
+ dprenet_dropout_rate: 0.5
193
+ postnet_dropout_rate: 0.5
194
+ transformer_enc_dropout_rate: 0.1
195
+ transformer_enc_positional_dropout_rate: 0.1
196
+ transformer_enc_attn_dropout_rate: 0.1
197
+ transformer_dec_dropout_rate: 0.1
198
+ transformer_dec_positional_dropout_rate: 0.1
199
+ transformer_dec_attn_dropout_rate: 0.1
200
+ transformer_enc_dec_attn_dropout_rate: 0.1
201
+ use_guided_attn_loss: true
202
+ num_heads_applied_guided_attn: 2
203
+ num_layers_applied_guided_attn: 2
204
+ modules_applied_guided_attn:
205
+ - encoder-decoder
206
+ guided_attn_loss_lambda: 10.0
207
+ pitch_extract: null
208
+ pitch_extract_conf: {}
209
+ pitch_normalize: null
210
+ pitch_normalize_conf: {}
211
+ energy_extract: null
212
+ energy_extract_conf: {}
213
+ energy_normalize: null
214
+ energy_normalize_conf: {}
215
+ required:
216
+ - output_dir
217
+ - token_list
218
+ distributed: true
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/backward_time.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/bce_loss.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/decoder_alpha.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/enc_dec_attn_loss.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/encoder_alpha.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/forward_time.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/iter_time.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/l1_loss.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/l2_loss.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/loss.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/lr_0.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/optim_step_time.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/images/train_time.png ADDED
Binary file
exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/train.loss.ave_5best.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7de91b4ed16a2b2925658fc73beef77181dd073ea7356078a95643304ea3f6f
3
+ size 132463677
meta.yaml ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
1
+ espnet: 0.8.0
2
+ files:
3
+ model_file: exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/train.loss.ave_5best.pth
4
+ python: "3.7.3 (default, Mar 27 2019, 22:11:17) \n[GCC 7.3.0]"
5
+ timestamp: 1600329812.363689
6
+ torch: 1.5.1
7
+ yaml_files:
8
+ train_config: exp/tts_train_transformer_raw_phn_jaconv_pyopenjtalk/config.yaml