add config
Browse files- config.json +257 -0
config.json
ADDED
@@ -0,0 +1,257 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"output_path": ".",
|
3 |
+
"logger_uri": null,
|
4 |
+
"run_name": "vits_Bhojpuri_Male_30hrs",
|
5 |
+
"project_name": null,
|
6 |
+
"run_description": "\ud83d\udc38Coqui trainer run.",
|
7 |
+
"print_step": 25,
|
8 |
+
"plot_step": 100,
|
9 |
+
"model_param_stats": false,
|
10 |
+
"wandb_entity": null,
|
11 |
+
"dashboard_logger": "tensorboard",
|
12 |
+
"log_model_step": null,
|
13 |
+
"save_step": 20000,
|
14 |
+
"save_n_checkpoints": 1000,
|
15 |
+
"save_checkpoints": true,
|
16 |
+
"save_all_best": false,
|
17 |
+
"save_best_after": 10000,
|
18 |
+
"target_loss": null,
|
19 |
+
"print_eval": true,
|
20 |
+
"test_delay_epochs": -1,
|
21 |
+
"run_eval": true,
|
22 |
+
"run_eval_steps": null,
|
23 |
+
"distributed_backend": "nccl",
|
24 |
+
"distributed_url": "tcp://localhost:54321",
|
25 |
+
"mixed_precision": true,
|
26 |
+
"epochs": 500,
|
27 |
+
"batch_size": 40,
|
28 |
+
"eval_batch_size": 16,
|
29 |
+
"grad_clip": [
|
30 |
+
1000,
|
31 |
+
1000
|
32 |
+
],
|
33 |
+
"scheduler_after_epoch": true,
|
34 |
+
"lr": 0.001,
|
35 |
+
"optimizer": "AdamW",
|
36 |
+
"optimizer_params": {
|
37 |
+
"betas": [
|
38 |
+
0.8,
|
39 |
+
0.99
|
40 |
+
],
|
41 |
+
"eps": 1e-09,
|
42 |
+
"weight_decay": 0.01
|
43 |
+
},
|
44 |
+
"lr_scheduler": null,
|
45 |
+
"lr_scheduler_params": {},
|
46 |
+
"use_grad_scaler": false,
|
47 |
+
"cudnn_enable": true,
|
48 |
+
"cudnn_deterministic": false,
|
49 |
+
"cudnn_benchmark": false,
|
50 |
+
"training_seed": 54321,
|
51 |
+
"model": "vits",
|
52 |
+
"num_loader_workers": 8,
|
53 |
+
"num_eval_loader_workers": 4,
|
54 |
+
"use_noise_augment": false,
|
55 |
+
"audio": {
|
56 |
+
"fft_size": 1024,
|
57 |
+
"sample_rate": 22050,
|
58 |
+
"win_length": 1024,
|
59 |
+
"hop_length": 256,
|
60 |
+
"num_mels": 80,
|
61 |
+
"mel_fmin": 0,
|
62 |
+
"mel_fmax": null
|
63 |
+
},
|
64 |
+
"use_phonemes": false,
|
65 |
+
"phonemizer": null,
|
66 |
+
"phoneme_language": "en-us",
|
67 |
+
"compute_input_seq_cache": true,
|
68 |
+
"text_cleaner": "multilingual_cleaners",
|
69 |
+
"enable_eos_bos_chars": false,
|
70 |
+
"test_sentences_file": "",
|
71 |
+
"phoneme_cache_path": "./phoneme_cache",
|
72 |
+
"characters": {
|
73 |
+
"characters_class": "TTS.tts.models.vits.VitsCharacters",
|
74 |
+
"vocab_dict": null,
|
75 |
+
"pad": "<PAD>",
|
76 |
+
"eos": "<EOS>",
|
77 |
+
"bos": "<BOS>",
|
78 |
+
"blank": "<BLNK>",
|
79 |
+
"characters": "\u091a.\u0947\u0910\u0925\u092e\u0959\u091d\u0906\u0949?\u092d \u092a\u0939\u0928\u093d\u091f\u0938\u0935\u0940\u091b\u0923\u0921\u091e\u0926\u094b\u0915\u0924\u0948\u0943\u095b\u0941\u095e\u092c\u0908\u0946\u094c\u0927\u090b\u093e\u0922\u0907\u093c\u0902\u0905\u0937\u0920\u095c\u0913\u092f,\u093f\u0930\u0901\u0914\u092b\u0909\u0916\u0911\u094d\u0932\u091c\u090f\u090a\u0917\u0936\u095d\u0919\u0918\u0942",
|
80 |
+
"punctuations": "!\u00a1'(),-.:;\u00bf? ",
|
81 |
+
"phonemes": null,
|
82 |
+
"is_unique": true,
|
83 |
+
"is_sorted": true
|
84 |
+
},
|
85 |
+
"add_blank": true,
|
86 |
+
"batch_group_size": 5,
|
87 |
+
"loss_masking": null,
|
88 |
+
"min_audio_len": 1,
|
89 |
+
"max_audio_len": Infinity,
|
90 |
+
"min_text_len": 1,
|
91 |
+
"max_text_len": Infinity,
|
92 |
+
"compute_f0": false,
|
93 |
+
"compute_energy": false,
|
94 |
+
"compute_linear_spec": true,
|
95 |
+
"precompute_num_workers": 0,
|
96 |
+
"start_by_longest": false,
|
97 |
+
"shuffle": false,
|
98 |
+
"drop_last": false,
|
99 |
+
"datasets": [
|
100 |
+
{
|
101 |
+
"formatter": "syspin",
|
102 |
+
"dataset_name": "",
|
103 |
+
"path": ".",
|
104 |
+
"meta_file_train": "../manifests/Bhojpuri_Male/30hrs.tsv",
|
105 |
+
"ignored_speakers": null,
|
106 |
+
"language": "",
|
107 |
+
"phonemizer": "",
|
108 |
+
"meta_file_val": "",
|
109 |
+
"meta_file_attn_mask": ""
|
110 |
+
}
|
111 |
+
],
|
112 |
+
"test_sentences": [
|
113 |
+
[
|
114 |
+
"\u090f\u0928\u094d\u091f\u094d\u0930\u093e\u092a\u0940 \u0915\u0902\u092a\u094d\u092f\u0942\u091f\u093f\u0902\u0917 \u092e\u0947\u0902 \u090f\u0928\u094d\u091f\u094d\u0930\u094b\u092a\u0940 \u090a \u0911\u092a\u0930\u0947\u091f\u093f\u0902\u0917 \u0938\u093f\u0938\u094d\u091f\u092e \u0939 \u091c\u0947 \u092a\u0947 \u0938\u0930\u093e \u0915\u094d\u0930\u093f\u092a\u094d\u091f\u094b\u0917\u094d\u0930\u093e\u092b\u093f\u0915 \u092b\u0902\u0915\u094d\u0936\u0928 \u0938\u092c \u0915\u093e\u092e \u0915\u0930\u0947 \u0932\u0947\u0902",
|
115 |
+
"Bhojpuri_Male",
|
116 |
+
null,
|
117 |
+
"bh"
|
118 |
+
]
|
119 |
+
],
|
120 |
+
"eval_split_max_size": null,
|
121 |
+
"eval_split_size": 0.01,
|
122 |
+
"use_speaker_weighted_sampler": false,
|
123 |
+
"speaker_weighted_sampler_alpha": 1.0,
|
124 |
+
"use_language_weighted_sampler": false,
|
125 |
+
"language_weighted_sampler_alpha": 1.0,
|
126 |
+
"use_length_weighted_sampler": false,
|
127 |
+
"length_weighted_sampler_alpha": 1.0,
|
128 |
+
"model_args": {
|
129 |
+
"num_chars": 86,
|
130 |
+
"out_channels": 513,
|
131 |
+
"spec_segment_size": 32,
|
132 |
+
"hidden_channels": 192,
|
133 |
+
"hidden_channels_ffn_text_encoder": 768,
|
134 |
+
"num_heads_text_encoder": 2,
|
135 |
+
"num_layers_text_encoder": 6,
|
136 |
+
"kernel_size_text_encoder": 3,
|
137 |
+
"dropout_p_text_encoder": 0.1,
|
138 |
+
"dropout_p_duration_predictor": 0.5,
|
139 |
+
"kernel_size_posterior_encoder": 5,
|
140 |
+
"dilation_rate_posterior_encoder": 1,
|
141 |
+
"num_layers_posterior_encoder": 16,
|
142 |
+
"kernel_size_flow": 5,
|
143 |
+
"dilation_rate_flow": 1,
|
144 |
+
"num_layers_flow": 4,
|
145 |
+
"resblock_type_decoder": "1",
|
146 |
+
"resblock_kernel_sizes_decoder": [
|
147 |
+
3,
|
148 |
+
7,
|
149 |
+
11
|
150 |
+
],
|
151 |
+
"resblock_dilation_sizes_decoder": [
|
152 |
+
[
|
153 |
+
1,
|
154 |
+
3,
|
155 |
+
5
|
156 |
+
],
|
157 |
+
[
|
158 |
+
1,
|
159 |
+
3,
|
160 |
+
5
|
161 |
+
],
|
162 |
+
[
|
163 |
+
1,
|
164 |
+
3,
|
165 |
+
5
|
166 |
+
]
|
167 |
+
],
|
168 |
+
"upsample_rates_decoder": [
|
169 |
+
8,
|
170 |
+
8,
|
171 |
+
2,
|
172 |
+
2
|
173 |
+
],
|
174 |
+
"upsample_initial_channel_decoder": 512,
|
175 |
+
"upsample_kernel_sizes_decoder": [
|
176 |
+
16,
|
177 |
+
16,
|
178 |
+
4,
|
179 |
+
4
|
180 |
+
],
|
181 |
+
"periods_multi_period_discriminator": [
|
182 |
+
2,
|
183 |
+
3,
|
184 |
+
5,
|
185 |
+
7,
|
186 |
+
11
|
187 |
+
],
|
188 |
+
"use_sdp": true,
|
189 |
+
"noise_scale": 1.0,
|
190 |
+
"inference_noise_scale": 0.667,
|
191 |
+
"length_scale": 1,
|
192 |
+
"noise_scale_dp": 1.0,
|
193 |
+
"inference_noise_scale_dp": 1.0,
|
194 |
+
"max_inference_len": null,
|
195 |
+
"init_discriminator": true,
|
196 |
+
"use_spectral_norm_disriminator": false,
|
197 |
+
"use_speaker_embedding": false,
|
198 |
+
"num_speakers": 0,
|
199 |
+
"speakers_file": null,
|
200 |
+
"d_vector_file": null,
|
201 |
+
"speaker_embedding_channels": 256,
|
202 |
+
"use_d_vector_file": false,
|
203 |
+
"d_vector_dim": 0,
|
204 |
+
"detach_dp_input": true,
|
205 |
+
"use_language_embedding": false,
|
206 |
+
"embedded_language_dim": 4,
|
207 |
+
"num_languages": 0,
|
208 |
+
"language_ids_file": null,
|
209 |
+
"use_speaker_encoder_as_loss": false,
|
210 |
+
"speaker_encoder_config_path": "",
|
211 |
+
"speaker_encoder_model_path": "",
|
212 |
+
"condition_dp_on_speaker": true,
|
213 |
+
"freeze_encoder": false,
|
214 |
+
"freeze_DP": false,
|
215 |
+
"freeze_PE": false,
|
216 |
+
"freeze_flow_decoder": false,
|
217 |
+
"freeze_waveform_decoder": false,
|
218 |
+
"encoder_sample_rate": null,
|
219 |
+
"interpolate_z": true,
|
220 |
+
"reinit_DP": false,
|
221 |
+
"reinit_text_encoder": false
|
222 |
+
},
|
223 |
+
"lr_gen": 0.0002,
|
224 |
+
"lr_disc": 0.0002,
|
225 |
+
"lr_scheduler_gen": "ExponentialLR",
|
226 |
+
"lr_scheduler_gen_params": {
|
227 |
+
"gamma": 0.999875,
|
228 |
+
"last_epoch": -1
|
229 |
+
},
|
230 |
+
"lr_scheduler_disc": "ExponentialLR",
|
231 |
+
"lr_scheduler_disc_params": {
|
232 |
+
"gamma": 0.999875,
|
233 |
+
"last_epoch": -1
|
234 |
+
},
|
235 |
+
"kl_loss_alpha": 1.0,
|
236 |
+
"disc_loss_alpha": 1.0,
|
237 |
+
"gen_loss_alpha": 1.0,
|
238 |
+
"feat_loss_alpha": 1.0,
|
239 |
+
"mel_loss_alpha": 45.0,
|
240 |
+
"dur_loss_alpha": 1.0,
|
241 |
+
"speaker_encoder_loss_alpha": 1.0,
|
242 |
+
"return_wav": true,
|
243 |
+
"use_weighted_sampler": false,
|
244 |
+
"weighted_sampler_attrs": {},
|
245 |
+
"weighted_sampler_multipliers": {},
|
246 |
+
"r": 1,
|
247 |
+
"num_speakers": 0,
|
248 |
+
"use_speaker_embedding": false,
|
249 |
+
"speakers_file": null,
|
250 |
+
"speaker_embedding_channels": 256,
|
251 |
+
"language_ids_file": null,
|
252 |
+
"use_language_embedding": false,
|
253 |
+
"use_d_vector_file": false,
|
254 |
+
"d_vector_file": null,
|
255 |
+
"d_vector_dim": 0,
|
256 |
+
"github_branch": "* dev"
|
257 |
+
}
|