Update README.md
Browse files
README.md
CHANGED
@@ -12,7 +12,7 @@ widget:
|
|
12 |
---
|
13 |
## RyanSpeech model (based on ESPnet2)
|
14 |
|
15 |
-
### `espnet/
|
16 |
This model was trained by Rohola Zandie using ryanspeech recipe in [espnet](https://github.com/espnet/espnet/).
|
17 |
|
18 |
|
@@ -24,12 +24,12 @@ You can download RyanSpeech dataset from [here](https://www.kaggle.com/datasets/
|
|
24 |
<details><summary>expand</summary>
|
25 |
|
26 |
```
|
27 |
-
config: conf/
|
28 |
print_config: false
|
29 |
log_level: INFO
|
30 |
dry_run: false
|
31 |
iterator_type: sequence
|
32 |
-
output_dir: exp/
|
33 |
ngpu: 1
|
34 |
seed: 0
|
35 |
num_workers: 1
|
@@ -48,7 +48,7 @@ cudnn_benchmark: false
|
|
48 |
cudnn_deterministic: true
|
49 |
collect_stats: false
|
50 |
write_collected_feats: false
|
51 |
-
max_epoch:
|
52 |
patience: null
|
53 |
val_scheduler_criterion:
|
54 |
- valid
|
@@ -68,7 +68,7 @@ keep_nbest_models: 5
|
|
68 |
grad_clip: 1.0
|
69 |
grad_clip_type: 2.0
|
70 |
grad_noise: false
|
71 |
-
accum_grad:
|
72 |
no_forward_run: false
|
73 |
resume: true
|
74 |
train_dtype: float32
|
@@ -79,14 +79,14 @@ pretrain_key: []
|
|
79 |
num_iters_per_epoch: 500
|
80 |
batch_size: 20
|
81 |
valid_batch_size: null
|
82 |
-
batch_bins:
|
83 |
valid_batch_bins: null
|
84 |
train_shape_file:
|
85 |
-
- exp/
|
86 |
-
- exp/
|
87 |
valid_shape_file:
|
88 |
-
- exp/
|
89 |
-
- exp/
|
90 |
batch_type: numel
|
91 |
valid_batch_type: null
|
92 |
fold_length:
|
@@ -102,9 +102,6 @@ train_data_path_and_name_and_type:
|
|
102 |
- - dump/raw/tr_no_dev/text
|
103 |
- text
|
104 |
- text
|
105 |
-
- - exp/tts_train_raw_phn_tacotron_g2p_en_no_space/decode_use_teacher_forcingtrue_train.loss.best//tr_no_dev/durations
|
106 |
-
- durations
|
107 |
-
- text_int
|
108 |
- - dump/raw/tr_no_dev/wav.scp
|
109 |
- speech
|
110 |
- sound
|
@@ -112,9 +109,6 @@ valid_data_path_and_name_and_type:
|
|
112 |
- - dump/raw/dev/text
|
113 |
- text
|
114 |
- text
|
115 |
-
- - exp/tts_train_raw_phn_tacotron_g2p_en_no_space/decode_use_teacher_forcingtrue_train.loss.best//dev/durations
|
116 |
-
- durations
|
117 |
-
- text_int
|
118 |
- - dump/raw/dev/wav.scp
|
119 |
- speech
|
120 |
- sound
|
@@ -124,11 +118,11 @@ max_cache_fd: 32
|
|
124 |
valid_max_cache_size: null
|
125 |
optim: adam
|
126 |
optim_conf:
|
127 |
-
lr:
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
token_list:
|
133 |
- <blank>
|
134 |
- <unk>
|
@@ -226,37 +220,40 @@ feats_extract_conf:
|
|
226 |
win_length: null
|
227 |
normalize: global_mvn
|
228 |
normalize_conf:
|
229 |
-
stats_file: exp/
|
230 |
-
tts:
|
231 |
tts_conf:
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
|
239 |
-
|
240 |
-
|
241 |
-
|
242 |
-
|
|
|
|
|
|
|
|
|
243 |
postnet_layers: 5
|
|
|
244 |
postnet_filts: 5
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
|
|
250 |
reduction_factor: 1
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
transformer_dec_dropout_rate: 0.1
|
258 |
-
transformer_dec_positional_dropout_rate: 0.1
|
259 |
-
transformer_dec_attn_dropout_rate: 0.1
|
260 |
pitch_extract: null
|
261 |
pitch_extract_conf: {}
|
262 |
pitch_normalize: null
|
@@ -269,6 +266,7 @@ required:
|
|
269 |
- output_dir
|
270 |
- token_list
|
271 |
distributed: false
|
|
|
272 |
```
|
273 |
|
274 |
</details>
|
|
|
12 |
---
|
13 |
## RyanSpeech model (based on ESPnet2)
|
14 |
|
15 |
+
### `espnet/english_male_ryanspeech_tacotron`
|
16 |
This model was trained by Rohola Zandie using ryanspeech recipe in [espnet](https://github.com/espnet/espnet/).
|
17 |
|
18 |
|
|
|
24 |
<details><summary>expand</summary>
|
25 |
|
26 |
```
|
27 |
+
config: conf/train.yaml
|
28 |
print_config: false
|
29 |
log_level: INFO
|
30 |
dry_run: false
|
31 |
iterator_type: sequence
|
32 |
+
output_dir: exp/tts_train_raw_phn_tacotron_g2p_en_no_space
|
33 |
ngpu: 1
|
34 |
seed: 0
|
35 |
num_workers: 1
|
|
|
48 |
cudnn_deterministic: true
|
49 |
collect_stats: false
|
50 |
write_collected_feats: false
|
51 |
+
max_epoch: 200
|
52 |
patience: null
|
53 |
val_scheduler_criterion:
|
54 |
- valid
|
|
|
68 |
grad_clip: 1.0
|
69 |
grad_clip_type: 2.0
|
70 |
grad_noise: false
|
71 |
+
accum_grad: 1
|
72 |
no_forward_run: false
|
73 |
resume: true
|
74 |
train_dtype: float32
|
|
|
79 |
num_iters_per_epoch: 500
|
80 |
batch_size: 20
|
81 |
valid_batch_size: null
|
82 |
+
batch_bins: 5120000
|
83 |
valid_batch_bins: null
|
84 |
train_shape_file:
|
85 |
+
- exp/tts_stats_raw_phn_tacotron_g2p_en_no_space/train/text_shape.phn
|
86 |
+
- exp/tts_stats_raw_phn_tacotron_g2p_en_no_space/train/speech_shape
|
87 |
valid_shape_file:
|
88 |
+
- exp/tts_stats_raw_phn_tacotron_g2p_en_no_space/valid/text_shape.phn
|
89 |
+
- exp/tts_stats_raw_phn_tacotron_g2p_en_no_space/valid/speech_shape
|
90 |
batch_type: numel
|
91 |
valid_batch_type: null
|
92 |
fold_length:
|
|
|
102 |
- - dump/raw/tr_no_dev/text
|
103 |
- text
|
104 |
- text
|
|
|
|
|
|
|
105 |
- - dump/raw/tr_no_dev/wav.scp
|
106 |
- speech
|
107 |
- sound
|
|
|
109 |
- - dump/raw/dev/text
|
110 |
- text
|
111 |
- text
|
|
|
|
|
|
|
112 |
- - dump/raw/dev/wav.scp
|
113 |
- speech
|
114 |
- sound
|
|
|
118 |
valid_max_cache_size: null
|
119 |
optim: adam
|
120 |
optim_conf:
|
121 |
+
lr: 0.001
|
122 |
+
eps: 1.0e-06
|
123 |
+
weight_decay: 0.0
|
124 |
+
scheduler: null
|
125 |
+
scheduler_conf: {}
|
126 |
token_list:
|
127 |
- <blank>
|
128 |
- <unk>
|
|
|
220 |
win_length: null
|
221 |
normalize: global_mvn
|
222 |
normalize_conf:
|
223 |
+
stats_file: exp/tts_stats_raw_phn_tacotron_g2p_en_no_space/train/feats_stats.npz
|
224 |
+
tts: tacotron2
|
225 |
tts_conf:
|
226 |
+
embed_dim: 512
|
227 |
+
elayers: 1
|
228 |
+
eunits: 512
|
229 |
+
econv_layers: 3
|
230 |
+
econv_chans: 512
|
231 |
+
econv_filts: 5
|
232 |
+
atype: location
|
233 |
+
adim: 512
|
234 |
+
aconv_chans: 32
|
235 |
+
aconv_filts: 15
|
236 |
+
cumulate_att_w: true
|
237 |
+
dlayers: 2
|
238 |
+
dunits: 1024
|
239 |
+
prenet_layers: 2
|
240 |
+
prenet_units: 256
|
241 |
postnet_layers: 5
|
242 |
+
postnet_chans: 512
|
243 |
postnet_filts: 5
|
244 |
+
output_activation: null
|
245 |
+
use_batch_norm: true
|
246 |
+
use_concate: true
|
247 |
+
use_residual: false
|
248 |
+
dropout_rate: 0.5
|
249 |
+
zoneout_rate: 0.1
|
250 |
reduction_factor: 1
|
251 |
+
spk_embed_dim: null
|
252 |
+
use_masking: true
|
253 |
+
bce_pos_weight: 5.0
|
254 |
+
use_guided_attn_loss: true
|
255 |
+
guided_attn_loss_sigma: 0.4
|
256 |
+
guided_attn_loss_lambda: 1.0
|
|
|
|
|
|
|
257 |
pitch_extract: null
|
258 |
pitch_extract_conf: {}
|
259 |
pitch_normalize: null
|
|
|
266 |
- output_dir
|
267 |
- token_list
|
268 |
distributed: false
|
269 |
+
|
270 |
```
|
271 |
|
272 |
</details>
|