Spaces:
Sleeping
Sleeping
kevinwang676
commited on
Commit
•
e738acb
1
Parent(s):
1083bf3
Upload folder using huggingface_hub
Browse files- finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model.pth +3 -0
- finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_282.pth +3 -0
- finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/config.json +159 -0
- finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/events.out.tfevents.1720170558.3a6c68407e6e.20750.0 +3 -0
- finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/trainer_0_log.txt +272 -0
- finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/vocab.json +0 -0
- finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/webui_xtts.py +1115 -0
- finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/config.json +159 -0
- finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/dvae.pth +3 -0
- finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/mel_stats.pth +3 -0
- finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/model.pth +3 -0
- finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/vocab.json +0 -0
finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2f627360fb46bb9093c3af9fa233264b5a430c1537771f199f669337161279d7
|
3 |
+
size 5607927061
|
finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_282.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2f627360fb46bb9093c3af9fa233264b5a430c1537771f199f669337161279d7
|
3 |
+
size 5607927061
|
finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/config.json
ADDED
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"output_path": "output",
|
3 |
+
"logger_uri": null,
|
4 |
+
"run_name": "run",
|
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 |
+
"save_on_interrupt": true,
|
13 |
+
"log_model_step": null,
|
14 |
+
"save_step": 10000,
|
15 |
+
"save_n_checkpoints": 5,
|
16 |
+
"save_checkpoints": true,
|
17 |
+
"save_all_best": false,
|
18 |
+
"save_best_after": 10000,
|
19 |
+
"target_loss": null,
|
20 |
+
"print_eval": false,
|
21 |
+
"test_delay_epochs": 0,
|
22 |
+
"run_eval": true,
|
23 |
+
"run_eval_steps": null,
|
24 |
+
"distributed_backend": "nccl",
|
25 |
+
"distributed_url": "tcp://localhost:54321",
|
26 |
+
"mixed_precision": false,
|
27 |
+
"precision": "fp16",
|
28 |
+
"epochs": 1000,
|
29 |
+
"batch_size": 32,
|
30 |
+
"eval_batch_size": 16,
|
31 |
+
"grad_clip": 0.0,
|
32 |
+
"scheduler_after_epoch": true,
|
33 |
+
"lr": 0.001,
|
34 |
+
"optimizer": "radam",
|
35 |
+
"optimizer_params": null,
|
36 |
+
"lr_scheduler": null,
|
37 |
+
"lr_scheduler_params": {},
|
38 |
+
"use_grad_scaler": false,
|
39 |
+
"allow_tf32": false,
|
40 |
+
"cudnn_enable": true,
|
41 |
+
"cudnn_deterministic": false,
|
42 |
+
"cudnn_benchmark": false,
|
43 |
+
"training_seed": 54321,
|
44 |
+
"model": "xtts",
|
45 |
+
"num_loader_workers": 0,
|
46 |
+
"num_eval_loader_workers": 0,
|
47 |
+
"use_noise_augment": false,
|
48 |
+
"audio": {
|
49 |
+
"sample_rate": 22050,
|
50 |
+
"output_sample_rate": 24000
|
51 |
+
},
|
52 |
+
"use_phonemes": false,
|
53 |
+
"phonemizer": null,
|
54 |
+
"phoneme_language": null,
|
55 |
+
"compute_input_seq_cache": false,
|
56 |
+
"text_cleaner": null,
|
57 |
+
"enable_eos_bos_chars": false,
|
58 |
+
"test_sentences_file": "",
|
59 |
+
"phoneme_cache_path": null,
|
60 |
+
"characters": null,
|
61 |
+
"add_blank": false,
|
62 |
+
"batch_group_size": 0,
|
63 |
+
"loss_masking": null,
|
64 |
+
"min_audio_len": 1,
|
65 |
+
"max_audio_len": Infinity,
|
66 |
+
"min_text_len": 1,
|
67 |
+
"max_text_len": Infinity,
|
68 |
+
"compute_f0": false,
|
69 |
+
"compute_energy": false,
|
70 |
+
"compute_linear_spec": false,
|
71 |
+
"precompute_num_workers": 0,
|
72 |
+
"start_by_longest": false,
|
73 |
+
"shuffle": false,
|
74 |
+
"drop_last": false,
|
75 |
+
"datasets": [
|
76 |
+
{
|
77 |
+
"formatter": "",
|
78 |
+
"dataset_name": "",
|
79 |
+
"path": "",
|
80 |
+
"meta_file_train": "",
|
81 |
+
"ignored_speakers": null,
|
82 |
+
"language": "",
|
83 |
+
"phonemizer": "",
|
84 |
+
"meta_file_val": "",
|
85 |
+
"meta_file_attn_mask": ""
|
86 |
+
}
|
87 |
+
],
|
88 |
+
"test_sentences": [],
|
89 |
+
"eval_split_max_size": null,
|
90 |
+
"eval_split_size": 0.01,
|
91 |
+
"use_speaker_weighted_sampler": false,
|
92 |
+
"speaker_weighted_sampler_alpha": 1.0,
|
93 |
+
"use_language_weighted_sampler": false,
|
94 |
+
"language_weighted_sampler_alpha": 1.0,
|
95 |
+
"use_length_weighted_sampler": false,
|
96 |
+
"length_weighted_sampler_alpha": 1.0,
|
97 |
+
"model_args": {
|
98 |
+
"gpt_batch_size": 1,
|
99 |
+
"enable_redaction": false,
|
100 |
+
"kv_cache": true,
|
101 |
+
"gpt_checkpoint": null,
|
102 |
+
"clvp_checkpoint": null,
|
103 |
+
"decoder_checkpoint": null,
|
104 |
+
"num_chars": 255,
|
105 |
+
"tokenizer_file": "",
|
106 |
+
"gpt_max_audio_tokens": 605,
|
107 |
+
"gpt_max_text_tokens": 402,
|
108 |
+
"gpt_max_prompt_tokens": 70,
|
109 |
+
"gpt_layers": 30,
|
110 |
+
"gpt_n_model_channels": 1024,
|
111 |
+
"gpt_n_heads": 16,
|
112 |
+
"gpt_number_text_tokens": 6681,
|
113 |
+
"gpt_start_text_token": null,
|
114 |
+
"gpt_stop_text_token": null,
|
115 |
+
"gpt_num_audio_tokens": 1026,
|
116 |
+
"gpt_start_audio_token": 1024,
|
117 |
+
"gpt_stop_audio_token": 1025,
|
118 |
+
"gpt_code_stride_len": 1024,
|
119 |
+
"gpt_use_masking_gt_prompt_approach": true,
|
120 |
+
"gpt_use_perceiver_resampler": true,
|
121 |
+
"input_sample_rate": 22050,
|
122 |
+
"output_sample_rate": 24000,
|
123 |
+
"output_hop_length": 256,
|
124 |
+
"decoder_input_dim": 1024,
|
125 |
+
"d_vector_dim": 512,
|
126 |
+
"cond_d_vector_in_each_upsampling_layer": true,
|
127 |
+
"duration_const": 102400
|
128 |
+
},
|
129 |
+
"model_dir": null,
|
130 |
+
"languages": [
|
131 |
+
"en",
|
132 |
+
"es",
|
133 |
+
"fr",
|
134 |
+
"de",
|
135 |
+
"it",
|
136 |
+
"pt",
|
137 |
+
"pl",
|
138 |
+
"tr",
|
139 |
+
"ru",
|
140 |
+
"nl",
|
141 |
+
"cs",
|
142 |
+
"ar",
|
143 |
+
"zh-cn",
|
144 |
+
"hu",
|
145 |
+
"ko",
|
146 |
+
"ja",
|
147 |
+
"hi"
|
148 |
+
],
|
149 |
+
"temperature": 0.75,
|
150 |
+
"length_penalty": 1.0,
|
151 |
+
"repetition_penalty": 5.0,
|
152 |
+
"top_k": 50,
|
153 |
+
"top_p": 0.85,
|
154 |
+
"num_gpt_outputs": 1,
|
155 |
+
"gpt_cond_len": 30,
|
156 |
+
"gpt_cond_chunk_len": 4,
|
157 |
+
"max_ref_len": 30,
|
158 |
+
"sound_norm_refs": false
|
159 |
+
}
|
finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/events.out.tfevents.1720170558.3a6c68407e6e.20750.0
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4adebfc072904a79436e20d51269031e9942096f1b14b3f34018574f26c70018
|
3 |
+
size 65761
|
finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/trainer_0_log.txt
ADDED
@@ -0,0 +1,272 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
> Training Environment:
|
2 |
+
| > Backend: Torch
|
3 |
+
| > Mixed precision: False
|
4 |
+
| > Precision: float32
|
5 |
+
| > Current device: 0
|
6 |
+
| > Num. of GPUs: 1
|
7 |
+
| > Num. of CPUs: 12
|
8 |
+
| > Num. of Torch Threads: 1
|
9 |
+
| > Torch seed: 1
|
10 |
+
| > Torch CUDNN: True
|
11 |
+
| > Torch CUDNN deterministic: False
|
12 |
+
| > Torch CUDNN benchmark: False
|
13 |
+
| > Torch TF32 MatMul: False
|
14 |
+
> Start Tensorboard: tensorboard --logdir=/content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
15 |
+
|
16 |
+
> Model has 518442047 parameters
|
17 |
+
|
18 |
+
[4m[1m > EPOCH: 0/10[0m
|
19 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
20 |
+
|
21 |
+
[1m > TRAINING (2024-07-05 09:09:19) [0m
|
22 |
+
|
23 |
+
[1m --> TIME: 2024-07-05 09:09:23 -- STEP: 0/47 -- GLOBAL_STEP: 0[0m
|
24 |
+
| > loss_text_ce: 0.021767420694231987 (0.021767420694231987)
|
25 |
+
| > loss_mel_ce: 4.51938533782959 (4.51938533782959)
|
26 |
+
| > loss: 4.5411529541015625 (4.5411529541015625)
|
27 |
+
| > grad_norm: 0 (0)
|
28 |
+
| > current_lr: 5e-06
|
29 |
+
| > step_time: 1.2336 (1.2336344718933105)
|
30 |
+
| > loader_time: 2.7081 (2.70807147026062)
|
31 |
+
|
32 |
+
|
33 |
+
[1m > EVALUATION [0m
|
34 |
+
|
35 |
+
|
36 |
+
[1m--> EVAL PERFORMANCE[0m
|
37 |
+
| > avg_loader_time: 0.14072567224502563 [0m(+0)
|
38 |
+
| > avg_loss_text_ce: 0.020066186785697937 [0m(+0)
|
39 |
+
| > avg_loss_mel_ce: 3.6496036052703857 [0m(+0)
|
40 |
+
| > avg_loss: 3.6696697771549225 [0m(+0)
|
41 |
+
|
42 |
+
> BEST MODEL : /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_47.pth
|
43 |
+
|
44 |
+
[4m[1m > EPOCH: 1/10[0m
|
45 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
46 |
+
|
47 |
+
[1m > TRAINING (2024-07-05 09:10:35) [0m
|
48 |
+
|
49 |
+
[1m --> TIME: 2024-07-05 09:10:40 -- STEP: 3/47 -- GLOBAL_STEP: 50[0m
|
50 |
+
| > loss_text_ce: 0.020956600084900856 (0.019851436217625935)
|
51 |
+
| > loss_mel_ce: 3.581146478652954 (3.554725726445516)
|
52 |
+
| > loss: 3.6021029949188232 (3.5745771725972495)
|
53 |
+
| > grad_norm: 0 (0.0)
|
54 |
+
| > current_lr: 5e-06
|
55 |
+
| > step_time: 0.4858 (0.5043480396270752)
|
56 |
+
| > loader_time: 0.0091 (0.01120901107788086)
|
57 |
+
|
58 |
+
|
59 |
+
[1m > EVALUATION [0m
|
60 |
+
|
61 |
+
|
62 |
+
[1m--> EVAL PERFORMANCE[0m
|
63 |
+
| > avg_loader_time:[92m 0.12269639968872069 [0m(-0.018029272556304946)
|
64 |
+
| > avg_loss_text_ce:[92m 0.01985420030541718 [0m(-0.00021198648028075695)
|
65 |
+
| > avg_loss_mel_ce:[92m 3.578354626893997 [0m(-0.07124897837638855)
|
66 |
+
| > avg_loss:[92m 3.5982088148593903 [0m(-0.07146096229553223)
|
67 |
+
|
68 |
+
> BEST MODEL : /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_94.pth
|
69 |
+
|
70 |
+
[4m[1m > EPOCH: 2/10[0m
|
71 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
72 |
+
|
73 |
+
[1m > TRAINING (2024-07-05 09:11:46) [0m
|
74 |
+
|
75 |
+
[1m --> TIME: 2024-07-05 09:11:53 -- STEP: 6/47 -- GLOBAL_STEP: 100[0m
|
76 |
+
| > loss_text_ce: 0.020052963867783546 (0.020112846046686172)
|
77 |
+
| > loss_mel_ce: 3.231581687927246 (3.4768107732137046)
|
78 |
+
| > loss: 3.2516345977783203 (3.496923565864563)
|
79 |
+
| > grad_norm: 0 (0.0)
|
80 |
+
| > current_lr: 5e-06
|
81 |
+
| > step_time: 0.4858 (0.44514886538187665)
|
82 |
+
| > loader_time: 0.0086 (0.016741156578063965)
|
83 |
+
|
84 |
+
|
85 |
+
[1m > EVALUATION [0m
|
86 |
+
|
87 |
+
|
88 |
+
[1m--> EVAL PERFORMANCE[0m
|
89 |
+
| > avg_loader_time:[92m 0.12173241376876831 [0m(-0.0009639859199523787)
|
90 |
+
| > avg_loss_text_ce:[92m 0.01967587019316852 [0m(-0.00017833011224865913)
|
91 |
+
| > avg_loss_mel_ce:[92m 3.5548452138900757 [0m(-0.02350941300392151)
|
92 |
+
| > avg_loss:[92m 3.574521094560623 [0m(-0.02368772029876709)
|
93 |
+
|
94 |
+
> BEST MODEL : /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_141.pth
|
95 |
+
|
96 |
+
[4m[1m > EPOCH: 3/10[0m
|
97 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
98 |
+
|
99 |
+
[1m > TRAINING (2024-07-05 09:12:57) [0m
|
100 |
+
|
101 |
+
[1m --> TIME: 2024-07-05 09:13:05 -- STEP: 9/47 -- GLOBAL_STEP: 150[0m
|
102 |
+
| > loss_text_ce: 0.020035386085510254 (0.02069739955994818)
|
103 |
+
| > loss_mel_ce: 3.9558589458465576 (3.411499950620863)
|
104 |
+
| > loss: 3.9758944511413574 (3.432197411855062)
|
105 |
+
| > grad_norm: 0 (0.0)
|
106 |
+
| > current_lr: 5e-06
|
107 |
+
| > step_time: 0.4835 (0.40245511796739364)
|
108 |
+
| > loader_time: 0.0315 (0.016773197385999892)
|
109 |
+
|
110 |
+
|
111 |
+
[1m > EVALUATION [0m
|
112 |
+
|
113 |
+
|
114 |
+
[1m--> EVAL PERFORMANCE[0m
|
115 |
+
| > avg_loader_time:[91m 0.12368443608283995 [0m(+0.0019520223140716414)
|
116 |
+
| > avg_loss_text_ce:[92m 0.01956212823279202 [0m(-0.00011374196037650108)
|
117 |
+
| > avg_loss_mel_ce:[92m 3.5282795131206512 [0m(-0.02656570076942444)
|
118 |
+
| > avg_loss:[92m 3.547841638326645 [0m(-0.02667945623397827)
|
119 |
+
|
120 |
+
> BEST MODEL : /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_188.pth
|
121 |
+
|
122 |
+
[4m[1m > EPOCH: 4/10[0m
|
123 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
124 |
+
|
125 |
+
[1m > TRAINING (2024-07-05 09:14:06) [0m
|
126 |
+
|
127 |
+
[1m --> TIME: 2024-07-05 09:14:17 -- STEP: 12/47 -- GLOBAL_STEP: 200[0m
|
128 |
+
| > loss_text_ce: 0.019074762240052223 (0.019689623111238085)
|
129 |
+
| > loss_mel_ce: 3.347362995147705 (3.35811048746109)
|
130 |
+
| > loss: 3.3664376735687256 (3.3778001070022583)
|
131 |
+
| > grad_norm: 0 (0.0)
|
132 |
+
| > current_lr: 5e-06
|
133 |
+
| > step_time: 0.3952 (0.4009103973706563)
|
134 |
+
| > loader_time: 0.0088 (0.009945313135782877)
|
135 |
+
|
136 |
+
|
137 |
+
[1m > EVALUATION [0m
|
138 |
+
|
139 |
+
|
140 |
+
[1m--> EVAL PERFORMANCE[0m
|
141 |
+
| > avg_loader_time:[91m 0.12455335259437561 [0m(+0.0008689165115356584)
|
142 |
+
| > avg_loss_text_ce:[92m 0.019442945485934615 [0m(-0.00011918274685740471)
|
143 |
+
| > avg_loss_mel_ce:[92m 3.514339804649353 [0m(-0.013939708471298218)
|
144 |
+
| > avg_loss:[92m 3.5337827503681183 [0m(-0.014058887958526611)
|
145 |
+
|
146 |
+
> BEST MODEL : /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_235.pth
|
147 |
+
|
148 |
+
[4m[1m > EPOCH: 5/10[0m
|
149 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
150 |
+
|
151 |
+
[1m > TRAINING (2024-07-05 09:15:13) [0m
|
152 |
+
|
153 |
+
[1m --> TIME: 2024-07-05 09:15:25 -- STEP: 15/47 -- GLOBAL_STEP: 250[0m
|
154 |
+
| > loss_text_ce: 0.019256995990872383 (0.020605798810720444)
|
155 |
+
| > loss_mel_ce: 2.906339406967163 (3.2246806462605795)
|
156 |
+
| > loss: 2.9255964756011963 (3.245286464691162)
|
157 |
+
| > grad_norm: 0 (0.0)
|
158 |
+
| > current_lr: 5e-06
|
159 |
+
| > step_time: 0.4036 (0.3896603584289551)
|
160 |
+
| > loader_time: 0.0085 (0.010488621393839518)
|
161 |
+
|
162 |
+
|
163 |
+
[1m > EVALUATION [0m
|
164 |
+
|
165 |
+
|
166 |
+
[1m--> EVAL PERFORMANCE[0m
|
167 |
+
| > avg_loader_time:[91m 0.126901775598526 [0m(+0.0023484230041503906)
|
168 |
+
| > avg_loss_text_ce:[92m 0.019381990423426032 [0m(-6.095506250858307e-05)
|
169 |
+
| > avg_loss_mel_ce:[92m 3.514001876115799 [0m(-0.00033792853355407715)
|
170 |
+
| > avg_loss:[92m 3.533383846282959 [0m(-0.00039890408515930176)
|
171 |
+
|
172 |
+
> BEST MODEL : /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/best_model_282.pth
|
173 |
+
|
174 |
+
[4m[1m > EPOCH: 6/10[0m
|
175 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
176 |
+
|
177 |
+
[1m > TRAINING (2024-07-05 09:16:28) [0m
|
178 |
+
|
179 |
+
[1m --> TIME: 2024-07-05 09:16:43 -- STEP: 18/47 -- GLOBAL_STEP: 300[0m
|
180 |
+
| > loss_text_ce: 0.019461730495095253 (0.020228115945226617)
|
181 |
+
| > loss_mel_ce: 3.325740337371826 (3.1609962781270347)
|
182 |
+
| > loss: 3.3452019691467285 (3.181224372651842)
|
183 |
+
| > grad_norm: 0 (0.0)
|
184 |
+
| > current_lr: 5e-06
|
185 |
+
| > step_time: 0.5458 (0.4343852864371406)
|
186 |
+
| > loader_time: 0.0126 (0.009873721334669325)
|
187 |
+
|
188 |
+
|
189 |
+
[1m > EVALUATION [0m
|
190 |
+
|
191 |
+
|
192 |
+
[1m--> EVAL PERFORMANCE[0m
|
193 |
+
| > avg_loader_time:[92m 0.12345081567764282 [0m(-0.0034509599208831787)
|
194 |
+
| > avg_loss_text_ce:[92m 0.01931103505194187 [0m(-7.095537148416042e-05)
|
195 |
+
| > avg_loss_mel_ce:[91m 3.5366199910640717 [0m(+0.022618114948272705)
|
196 |
+
| > avg_loss:[91m 3.5559310615062714 [0m(+0.022547215223312378)
|
197 |
+
|
198 |
+
|
199 |
+
[4m[1m > EPOCH: 7/10[0m
|
200 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
201 |
+
|
202 |
+
[1m > TRAINING (2024-07-05 09:17:06) [0m
|
203 |
+
|
204 |
+
[1m --> TIME: 2024-07-05 09:17:23 -- STEP: 21/47 -- GLOBAL_STEP: 350[0m
|
205 |
+
| > loss_text_ce: 0.021202489733695984 (0.020701753241675242)
|
206 |
+
| > loss_mel_ce: 3.258589029312134 (3.072263036455427)
|
207 |
+
| > loss: 3.2797915935516357 (3.0929648194994246)
|
208 |
+
| > grad_norm: 0 (0.0)
|
209 |
+
| > current_lr: 5e-06
|
210 |
+
| > step_time: 0.3435 (0.3992774713607061)
|
211 |
+
| > loader_time: 0.0094 (0.011507170540945872)
|
212 |
+
|
213 |
+
|
214 |
+
[1m > EVALUATION [0m
|
215 |
+
|
216 |
+
|
217 |
+
[1m--> EVAL PERFORMANCE[0m
|
218 |
+
| > avg_loader_time:[91m 0.12718188762664795 [0m(+0.003731071949005127)
|
219 |
+
| > avg_loss_text_ce:[92m 0.01926761632785201 [0m(-4.3418724089860916e-05)
|
220 |
+
| > avg_loss_mel_ce:[91m 3.539694160223007 [0m(+0.003074169158935547)
|
221 |
+
| > avg_loss:[91m 3.5589617490768433 [0m(+0.0030306875705718994)
|
222 |
+
|
223 |
+
|
224 |
+
[4m[1m > EPOCH: 8/10[0m
|
225 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
226 |
+
|
227 |
+
[1m > TRAINING (2024-07-05 09:17:44) [0m
|
228 |
+
|
229 |
+
[1m --> TIME: 2024-07-05 09:18:03 -- STEP: 24/47 -- GLOBAL_STEP: 400[0m
|
230 |
+
| > loss_text_ce: 0.02007140778005123 (0.020090758102014664)
|
231 |
+
| > loss_mel_ce: 3.1900110244750977 (2.9764333168665567)
|
232 |
+
| > loss: 3.210082530975342 (2.9965241154034934)
|
233 |
+
| > grad_norm: 0 (0.0)
|
234 |
+
| > current_lr: 5e-06
|
235 |
+
| > step_time: 0.4834 (0.4396652082602183)
|
236 |
+
| > loader_time: 0.01 (0.01113287607828776)
|
237 |
+
|
238 |
+
|
239 |
+
[1m > EVALUATION [0m
|
240 |
+
|
241 |
+
|
242 |
+
[1m--> EVAL PERFORMANCE[0m
|
243 |
+
| > avg_loader_time:[91m 0.12738469243049622 [0m(+0.0002028048038482666)
|
244 |
+
| > avg_loss_text_ce:[92m 0.01921596471220255 [0m(-5.1651615649461746e-05)
|
245 |
+
| > avg_loss_mel_ce:[91m 3.5459631085395813 [0m(+0.006268948316574097)
|
246 |
+
| > avg_loss:[91m 3.565179079771042 [0m(+0.006217330694198608)
|
247 |
+
|
248 |
+
|
249 |
+
[4m[1m > EPOCH: 9/10[0m
|
250 |
+
--> /content/xtts-v2/finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9
|
251 |
+
|
252 |
+
[1m > TRAINING (2024-07-05 09:18:22) [0m
|
253 |
+
|
254 |
+
[1m --> TIME: 2024-07-05 09:18:43 -- STEP: 27/47 -- GLOBAL_STEP: 450[0m
|
255 |
+
| > loss_text_ce: 0.019915737211704254 (0.02003813838517225)
|
256 |
+
| > loss_mel_ce: 3.058342933654785 (2.935793735362865)
|
257 |
+
| > loss: 3.078258752822876 (2.955831872092353)
|
258 |
+
| > grad_norm: 0 (0.0)
|
259 |
+
| > current_lr: 5e-06
|
260 |
+
| > step_time: 0.4082 (0.4022120723017939)
|
261 |
+
| > loader_time: 0.01 (0.01060099954958315)
|
262 |
+
|
263 |
+
|
264 |
+
[1m > EVALUATION [0m
|
265 |
+
|
266 |
+
|
267 |
+
[1m--> EVAL PERFORMANCE[0m
|
268 |
+
| > avg_loader_time:[91m 0.12931856513023376 [0m(+0.0019338726997375488)
|
269 |
+
| > avg_loss_text_ce:[92m 0.01917076646350324 [0m(-4.519824869930744e-05)
|
270 |
+
| > avg_loss_mel_ce:[91m 3.5662526190280914 [0m(+0.020289510488510132)
|
271 |
+
| > avg_loss:[91m 3.5854234099388123 [0m(+0.020244330167770386)
|
272 |
+
|
finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/vocab.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
finetune_models_kobe/run/training/GPT_XTTS_FT-July-05-2024_09+09AM-44c61c9/webui_xtts.py
ADDED
@@ -0,0 +1,1115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os,shutil,sys,pdb,re
|
2 |
+
now_dir = os.getcwd()
|
3 |
+
sys.path.append(now_dir)
|
4 |
+
import json,yaml,warnings,torch
|
5 |
+
import platform
|
6 |
+
import psutil
|
7 |
+
import signal
|
8 |
+
from pathlib import Path
|
9 |
+
|
10 |
+
warnings.filterwarnings("ignore")
|
11 |
+
torch.manual_seed(233333)
|
12 |
+
tmp = os.path.join(now_dir, "TEMP")
|
13 |
+
os.makedirs(tmp, exist_ok=True)
|
14 |
+
os.environ["TEMP"] = tmp
|
15 |
+
if(os.path.exists(tmp)):
|
16 |
+
for name in os.listdir(tmp):
|
17 |
+
if(name=="jieba.cache"):continue
|
18 |
+
path="%s/%s"%(tmp,name)
|
19 |
+
delete=os.remove if os.path.isfile(path) else shutil.rmtree
|
20 |
+
try:
|
21 |
+
delete(path)
|
22 |
+
except Exception as e:
|
23 |
+
print(str(e))
|
24 |
+
pass
|
25 |
+
import site
|
26 |
+
site_packages_roots = []
|
27 |
+
for path in site.getsitepackages():
|
28 |
+
if "packages" in path:
|
29 |
+
site_packages_roots.append(path)
|
30 |
+
if(site_packages_roots==[]):site_packages_roots=["%s/runtime/Lib/site-packages" % now_dir]
|
31 |
+
#os.environ["OPENBLAS_NUM_THREADS"] = "4"
|
32 |
+
os.environ["no_proxy"] = "localhost, 127.0.0.1, ::1"
|
33 |
+
os.environ["all_proxy"] = ""
|
34 |
+
for site_packages_root in site_packages_roots:
|
35 |
+
if os.path.exists(site_packages_root):
|
36 |
+
try:
|
37 |
+
with open("%s/users.pth" % (site_packages_root), "w") as f:
|
38 |
+
f.write(
|
39 |
+
"%s\n%s/tools\n%s/tools/damo_asr\n%s/GPT_SoVITS\n%s/tools/uvr5"
|
40 |
+
% (now_dir, now_dir, now_dir, now_dir, now_dir)
|
41 |
+
)
|
42 |
+
break
|
43 |
+
except PermissionError:
|
44 |
+
pass
|
45 |
+
from tools import my_utils
|
46 |
+
import traceback
|
47 |
+
import shutil
|
48 |
+
import pdb
|
49 |
+
import gradio as gr
|
50 |
+
from subprocess import Popen
|
51 |
+
import signal
|
52 |
+
from config import python_exec,infer_device,is_half,exp_root,webui_port_main,webui_port_infer_tts,webui_port_uvr5,webui_port_subfix,is_share
|
53 |
+
from tools.i18n.i18n import I18nAuto
|
54 |
+
i18n = I18nAuto()
|
55 |
+
from scipy.io import wavfile
|
56 |
+
from tools.my_utils import load_audio
|
57 |
+
from multiprocessing import cpu_count
|
58 |
+
|
59 |
+
import argparse
|
60 |
+
import os
|
61 |
+
import sys
|
62 |
+
import tempfile
|
63 |
+
|
64 |
+
import gradio as gr
|
65 |
+
import librosa.display
|
66 |
+
import numpy as np
|
67 |
+
|
68 |
+
import torch
|
69 |
+
import torchaudio
|
70 |
+
import traceback
|
71 |
+
from TTS.demos.xtts_ft_demo.utils.formatter import format_audio_list
|
72 |
+
from TTS.demos.xtts_ft_demo.utils.gpt_train import train_gpt
|
73 |
+
|
74 |
+
from TTS.tts.configs.xtts_config import XttsConfig
|
75 |
+
from TTS.tts.models.xtts import Xtts
|
76 |
+
|
77 |
+
# from .list to .csv
|
78 |
+
import pandas as pd
|
79 |
+
from sklearn.model_selection import train_test_split
|
80 |
+
|
81 |
+
def split_csv(input_csv, train_csv, eval_csv, eval_size=0.15):
|
82 |
+
# Load the data from the CSV file
|
83 |
+
data = pd.read_csv(input_csv, delimiter='|', header=0)
|
84 |
+
|
85 |
+
# Split the data into training and evaluation sets
|
86 |
+
train_data, eval_data = train_test_split(data, test_size=eval_size, random_state=42)
|
87 |
+
|
88 |
+
# Save the training data to a CSV file
|
89 |
+
train_data.to_csv(train_csv, index=False, sep='|')
|
90 |
+
|
91 |
+
# Save the evaluation data to a CSV file
|
92 |
+
eval_data.to_csv(eval_csv, index=False, sep='|')
|
93 |
+
|
94 |
+
print("CSV files have been successfully split.")
|
95 |
+
|
96 |
+
|
97 |
+
def convert_list_to_csv(input_file, output_file):
|
98 |
+
try:
|
99 |
+
# Open the input .list file to read
|
100 |
+
with open(input_file, 'r', encoding='utf-8') as infile:
|
101 |
+
# Open the output .csv file to write
|
102 |
+
with open(output_file, 'w', encoding='utf-8') as outfile:
|
103 |
+
# Write the header to the CSV
|
104 |
+
outfile.write("audio_file|text|speaker_name\n")
|
105 |
+
# Process each line in the input file
|
106 |
+
for line in infile:
|
107 |
+
parts = line.strip().split('|')
|
108 |
+
if len(parts) == 4:
|
109 |
+
# Extract relevant parts: WAV file path and transcription
|
110 |
+
wav_path = parts[0]
|
111 |
+
transcription = parts[3]
|
112 |
+
# Write the formatted line to the CSV file
|
113 |
+
outfile.write(f"{wav_path}|{transcription}|coqui\n")
|
114 |
+
print("Conversion to CSV completed successfully.")
|
115 |
+
split_csv(output_file, "train.csv", "eval.csv")
|
116 |
+
print("Split completed successfully")
|
117 |
+
return "train.csv", "eval.csv"
|
118 |
+
except Exception as e:
|
119 |
+
print(f"An error occurred: {e}")
|
120 |
+
|
121 |
+
|
122 |
+
def clear_gpu_cache():
|
123 |
+
# clear the GPU cache
|
124 |
+
if torch.cuda.is_available():
|
125 |
+
torch.cuda.empty_cache()
|
126 |
+
|
127 |
+
XTTS_MODEL = None
|
128 |
+
def load_model(xtts_checkpoint, xtts_config, xtts_vocab):
|
129 |
+
global XTTS_MODEL
|
130 |
+
clear_gpu_cache()
|
131 |
+
if not xtts_checkpoint or not xtts_config or not xtts_vocab:
|
132 |
+
return "You need to run the previous steps or manually set the `XTTS checkpoint path`, `XTTS config path`, and `XTTS vocab path` fields !!"
|
133 |
+
config = XttsConfig()
|
134 |
+
config.load_json(xtts_config)
|
135 |
+
XTTS_MODEL = Xtts.init_from_config(config)
|
136 |
+
print("Loading XTTS model! ")
|
137 |
+
XTTS_MODEL.load_checkpoint(config, checkpoint_path=xtts_checkpoint, vocab_path=xtts_vocab, use_deepspeed=False)
|
138 |
+
if torch.cuda.is_available():
|
139 |
+
XTTS_MODEL.cuda()
|
140 |
+
|
141 |
+
print("模型已成功加载!")
|
142 |
+
return "模型已成功加载!"
|
143 |
+
|
144 |
+
def run_tts(lang, tts_text, speaker_audio_file):
|
145 |
+
if XTTS_MODEL is None or not speaker_audio_file:
|
146 |
+
return "您需要先执行第5步 - 加载模型", None, None
|
147 |
+
|
148 |
+
speaker_audio_file = "".join([item for item in speaker_audio_file.strip().split("\n") if item != ""])
|
149 |
+
gpt_cond_latent, speaker_embedding = XTTS_MODEL.get_conditioning_latents(audio_path=speaker_audio_file, gpt_cond_len=XTTS_MODEL.config.gpt_cond_len, max_ref_length=XTTS_MODEL.config.max_ref_len, sound_norm_refs=XTTS_MODEL.config.sound_norm_refs)
|
150 |
+
out = XTTS_MODEL.inference(
|
151 |
+
text=tts_text.strip(),
|
152 |
+
language=lang,
|
153 |
+
gpt_cond_latent=gpt_cond_latent,
|
154 |
+
speaker_embedding=speaker_embedding,
|
155 |
+
temperature=XTTS_MODEL.config.temperature, # Add custom parameters here
|
156 |
+
length_penalty=XTTS_MODEL.config.length_penalty,
|
157 |
+
repetition_penalty=XTTS_MODEL.config.repetition_penalty,
|
158 |
+
top_k=XTTS_MODEL.config.top_k,
|
159 |
+
top_p=XTTS_MODEL.config.top_p,
|
160 |
+
)
|
161 |
+
|
162 |
+
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
|
163 |
+
out["wav"] = torch.tensor(out["wav"]).unsqueeze(0)
|
164 |
+
out_path = fp.name
|
165 |
+
torchaudio.save(out_path, out["wav"], 24000)
|
166 |
+
|
167 |
+
return "推理成功,快来听听吧!", out_path, speaker_audio_file
|
168 |
+
|
169 |
+
|
170 |
+
|
171 |
+
|
172 |
+
# define a logger to redirect
|
173 |
+
class Logger:
|
174 |
+
def __init__(self, filename="log.out"):
|
175 |
+
self.log_file = filename
|
176 |
+
self.terminal = sys.stdout
|
177 |
+
self.log = open(self.log_file, "w")
|
178 |
+
|
179 |
+
def write(self, message):
|
180 |
+
self.terminal.write(message)
|
181 |
+
self.log.write(message)
|
182 |
+
|
183 |
+
def flush(self):
|
184 |
+
self.terminal.flush()
|
185 |
+
self.log.flush()
|
186 |
+
|
187 |
+
def isatty(self):
|
188 |
+
return False
|
189 |
+
|
190 |
+
# redirect stdout and stderr to a file
|
191 |
+
sys.stdout = Logger()
|
192 |
+
sys.stderr = sys.stdout
|
193 |
+
|
194 |
+
|
195 |
+
# logging.basicConfig(stream=sys.stdout, level=logging.INFO)
|
196 |
+
import logging
|
197 |
+
logging.basicConfig(
|
198 |
+
level=logging.WARNING,
|
199 |
+
format="%(asctime)s [%(levelname)s] %(message)s",
|
200 |
+
handlers=[
|
201 |
+
logging.StreamHandler(sys.stdout)
|
202 |
+
]
|
203 |
+
)
|
204 |
+
|
205 |
+
def read_logs():
|
206 |
+
sys.stdout.flush()
|
207 |
+
with open(sys.stdout.log_file, "r") as f:
|
208 |
+
return f.read()
|
209 |
+
|
210 |
+
|
211 |
+
os.environ['PYTORCH_ENABLE_MPS_FALLBACK'] = '1' # 当遇到mps不支持的步骤时使用cpu
|
212 |
+
|
213 |
+
n_cpu=cpu_count()
|
214 |
+
|
215 |
+
ngpu = torch.cuda.device_count()
|
216 |
+
gpu_infos = []
|
217 |
+
mem = []
|
218 |
+
if_gpu_ok = False
|
219 |
+
|
220 |
+
# 判断是否有能用来训练和加速推理的N卡
|
221 |
+
if torch.cuda.is_available() or ngpu != 0:
|
222 |
+
for i in range(ngpu):
|
223 |
+
gpu_name = torch.cuda.get_device_name(i)
|
224 |
+
if any(value in gpu_name.upper()for value in ["10","16","20","30","40","A2","A3","A4","P4","A50","500","A60","70","80","90","M4","T4","TITAN","L4","4060"]):
|
225 |
+
# A10#A100#V100#A40#P40#M40#K80#A4500
|
226 |
+
if_gpu_ok = True # 至少有一张能用的N卡
|
227 |
+
gpu_infos.append("%s\t%s" % (i, gpu_name))
|
228 |
+
mem.append(int(torch.cuda.get_device_properties(i).total_memory/ 1024/ 1024/ 1024+ 0.4))
|
229 |
+
# 判断是否支持mps加速
|
230 |
+
if torch.backends.mps.is_available():
|
231 |
+
if_gpu_ok = True
|
232 |
+
gpu_infos.append("%s\t%s" % ("0", "Apple GPU"))
|
233 |
+
mem.append(psutil.virtual_memory().total/ 1024 / 1024 / 1024) # 实测使用系统内存作为显存不会爆显存
|
234 |
+
|
235 |
+
if if_gpu_ok and len(gpu_infos) > 0:
|
236 |
+
gpu_info = "\n".join(gpu_infos)
|
237 |
+
default_batch_size = min(mem) // 2
|
238 |
+
else:
|
239 |
+
gpu_info = i18n("很遗憾您这没有能用的显卡来支持您训练")
|
240 |
+
default_batch_size = 1
|
241 |
+
gpus = "-".join([i[0] for i in gpu_infos])
|
242 |
+
|
243 |
+
pretrained_sovits_name="GPT_SoVITS/pretrained_models/s2G488k.pth"
|
244 |
+
pretrained_gpt_name="GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"
|
245 |
+
def get_weights_names():
|
246 |
+
SoVITS_names = [pretrained_sovits_name]
|
247 |
+
for name in os.listdir(SoVITS_weight_root):
|
248 |
+
if name.endswith(".pth"):SoVITS_names.append(name)
|
249 |
+
GPT_names = [pretrained_gpt_name]
|
250 |
+
for name in os.listdir(GPT_weight_root):
|
251 |
+
if name.endswith(".ckpt"): GPT_names.append(name)
|
252 |
+
return SoVITS_names,GPT_names
|
253 |
+
SoVITS_weight_root="SoVITS_weights"
|
254 |
+
GPT_weight_root="GPT_weights"
|
255 |
+
os.makedirs(SoVITS_weight_root,exist_ok=True)
|
256 |
+
os.makedirs(GPT_weight_root,exist_ok=True)
|
257 |
+
SoVITS_names,GPT_names = get_weights_names()
|
258 |
+
|
259 |
+
def custom_sort_key(s):
|
260 |
+
# 使用正则表达式提取字符串中的数字部分和非数字部分
|
261 |
+
parts = re.split('(\d+)', s)
|
262 |
+
# 将数字部分转换为整数,非数字部分保持不变
|
263 |
+
parts = [int(part) if part.isdigit() else part for part in parts]
|
264 |
+
return parts
|
265 |
+
|
266 |
+
def change_choices():
|
267 |
+
SoVITS_names, GPT_names = get_weights_names()
|
268 |
+
return {"choices": sorted(SoVITS_names,key=custom_sort_key), "__type__": "update"}, {"choices": sorted(GPT_names,key=custom_sort_key), "__type__": "update"}
|
269 |
+
|
270 |
+
p_label=None
|
271 |
+
p_uvr5=None
|
272 |
+
p_asr=None
|
273 |
+
p_denoise=None
|
274 |
+
p_tts_inference=None
|
275 |
+
|
276 |
+
def kill_proc_tree(pid, including_parent=True):
|
277 |
+
try:
|
278 |
+
parent = psutil.Process(pid)
|
279 |
+
except psutil.NoSuchProcess:
|
280 |
+
# Process already terminated
|
281 |
+
return
|
282 |
+
|
283 |
+
children = parent.children(recursive=True)
|
284 |
+
for child in children:
|
285 |
+
try:
|
286 |
+
os.kill(child.pid, signal.SIGTERM) # or signal.SIGKILL
|
287 |
+
except OSError:
|
288 |
+
pass
|
289 |
+
if including_parent:
|
290 |
+
try:
|
291 |
+
os.kill(parent.pid, signal.SIGTERM) # or signal.SIGKILL
|
292 |
+
except OSError:
|
293 |
+
pass
|
294 |
+
|
295 |
+
system=platform.system()
|
296 |
+
def kill_process(pid):
|
297 |
+
if(system=="Windows"):
|
298 |
+
cmd = "taskkill /t /f /pid %s" % pid
|
299 |
+
os.system(cmd)
|
300 |
+
else:
|
301 |
+
kill_proc_tree(pid)
|
302 |
+
|
303 |
+
|
304 |
+
def change_label(if_label,path_list):
|
305 |
+
global p_label
|
306 |
+
if(if_label==True and p_label==None):
|
307 |
+
path_list=my_utils.clean_path(path_list)
|
308 |
+
cmd = '"%s" tools/subfix_webui.py --load_list "%s" --webui_port %s --is_share %s'%(python_exec,path_list,webui_port_subfix,is_share)
|
309 |
+
yield i18n("打标工具WebUI已开启")
|
310 |
+
print(cmd)
|
311 |
+
p_label = Popen(cmd, shell=True)
|
312 |
+
elif(if_label==False and p_label!=None):
|
313 |
+
kill_process(p_label.pid)
|
314 |
+
p_label=None
|
315 |
+
yield i18n("打标工具WebUI已关闭")
|
316 |
+
|
317 |
+
def change_uvr5(if_uvr5):
|
318 |
+
global p_uvr5
|
319 |
+
if(if_uvr5==True and p_uvr5==None):
|
320 |
+
cmd = '"%s" tools/uvr5/webui.py "%s" %s %s %s'%(python_exec,infer_device,is_half,webui_port_uvr5,is_share)
|
321 |
+
yield i18n("UVR5已开启")
|
322 |
+
print(cmd)
|
323 |
+
p_uvr5 = Popen(cmd, shell=True)
|
324 |
+
elif(if_uvr5==False and p_uvr5!=None):
|
325 |
+
kill_process(p_uvr5.pid)
|
326 |
+
p_uvr5=None
|
327 |
+
yield i18n("UVR5已关闭")
|
328 |
+
|
329 |
+
def change_tts_inference(if_tts,bert_path,cnhubert_base_path,gpu_number,gpt_path,sovits_path):
|
330 |
+
global p_tts_inference
|
331 |
+
if(if_tts==True and p_tts_inference==None):
|
332 |
+
os.environ["gpt_path"]=gpt_path if "/" in gpt_path else "%s/%s"%(GPT_weight_root,gpt_path)
|
333 |
+
os.environ["sovits_path"]=sovits_path if "/"in sovits_path else "%s/%s"%(SoVITS_weight_root,sovits_path)
|
334 |
+
os.environ["cnhubert_base_path"]=cnhubert_base_path
|
335 |
+
os.environ["bert_path"]=bert_path
|
336 |
+
os.environ["_CUDA_VISIBLE_DEVICES"]=gpu_number
|
337 |
+
os.environ["is_half"]=str(is_half)
|
338 |
+
os.environ["infer_ttswebui"]=str(webui_port_infer_tts)
|
339 |
+
os.environ["is_share"]=str(is_share)
|
340 |
+
cmd = '"%s" GPT_SoVITS/inference_webui.py'%(python_exec)
|
341 |
+
yield i18n("TTS推理进程已开启")
|
342 |
+
print(cmd)
|
343 |
+
p_tts_inference = Popen(cmd, shell=True)
|
344 |
+
elif(if_tts==False and p_tts_inference!=None):
|
345 |
+
kill_process(p_tts_inference.pid)
|
346 |
+
p_tts_inference=None
|
347 |
+
yield i18n("TTS推理进程已关闭")
|
348 |
+
|
349 |
+
from tools.asr.config import asr_dict
|
350 |
+
def open_asr(asr_inp_dir, asr_opt_dir, asr_model, asr_model_size, asr_lang):
|
351 |
+
global p_asr
|
352 |
+
if(p_asr==None):
|
353 |
+
asr_inp_dir=my_utils.clean_path(asr_inp_dir)
|
354 |
+
cmd = f'"{python_exec}" tools/asr/{asr_dict[asr_model]["path"]}'
|
355 |
+
cmd += f' -i "{asr_inp_dir}"'
|
356 |
+
cmd += f' -o "{asr_opt_dir}"'
|
357 |
+
cmd += f' -s {asr_model_size}'
|
358 |
+
cmd += f' -l {asr_lang}'
|
359 |
+
cmd += " -p %s"%("float16"if is_half==True else "float32")
|
360 |
+
|
361 |
+
yield "ASR任务开启:%s"%cmd,{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
362 |
+
print(cmd)
|
363 |
+
p_asr = Popen(cmd, shell=True)
|
364 |
+
p_asr.wait()
|
365 |
+
p_asr=None
|
366 |
+
yield f"ASR任务完成, 查看终端进行下一步",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
367 |
+
else:
|
368 |
+
yield "已有正在进行的ASR任务,需先终止才能开启下一次任务",{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
369 |
+
# return None
|
370 |
+
|
371 |
+
def close_asr():
|
372 |
+
global p_asr
|
373 |
+
if(p_asr!=None):
|
374 |
+
kill_process(p_asr.pid)
|
375 |
+
p_asr=None
|
376 |
+
return "已终止ASR进程",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
377 |
+
def open_denoise(denoise_inp_dir, denoise_opt_dir):
|
378 |
+
global p_denoise
|
379 |
+
if(p_denoise==None):
|
380 |
+
denoise_inp_dir=my_utils.clean_path(denoise_inp_dir)
|
381 |
+
denoise_opt_dir=my_utils.clean_path(denoise_opt_dir)
|
382 |
+
cmd = '"%s" tools/cmd-denoise.py -i "%s" -o "%s" -p %s'%(python_exec,denoise_inp_dir,denoise_opt_dir,"float16"if is_half==True else "float32")
|
383 |
+
|
384 |
+
yield "语音降噪任务开启:%s"%cmd,{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
385 |
+
print(cmd)
|
386 |
+
p_denoise = Popen(cmd, shell=True)
|
387 |
+
p_denoise.wait()
|
388 |
+
p_denoise=None
|
389 |
+
yield f"语音降噪任务完成, 查看终端进行下一步",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
390 |
+
else:
|
391 |
+
yield "已有正在进行的语音降噪任务,需先终止才能开启下一次任务",{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
392 |
+
# return None
|
393 |
+
|
394 |
+
def close_denoise():
|
395 |
+
global p_denoise
|
396 |
+
if(p_denoise!=None):
|
397 |
+
kill_process(p_denoise.pid)
|
398 |
+
p_denoise=None
|
399 |
+
return "已终止语音降噪进程",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
400 |
+
|
401 |
+
p_train_SoVITS=None
|
402 |
+
def open1Ba(batch_size,total_epoch,exp_name,text_low_lr_rate,if_save_latest,if_save_every_weights,save_every_epoch,gpu_numbers1Ba,pretrained_s2G,pretrained_s2D):
|
403 |
+
global p_train_SoVITS
|
404 |
+
if(p_train_SoVITS==None):
|
405 |
+
with open("GPT_SoVITS/configs/s2.json")as f:
|
406 |
+
data=f.read()
|
407 |
+
data=json.loads(data)
|
408 |
+
s2_dir="%s/%s"%(exp_root,exp_name)
|
409 |
+
os.makedirs("%s/logs_s2"%(s2_dir),exist_ok=True)
|
410 |
+
if(is_half==False):
|
411 |
+
data["train"]["fp16_run"]=False
|
412 |
+
batch_size=max(1,batch_size//2)
|
413 |
+
data["train"]["batch_size"]=batch_size
|
414 |
+
data["train"]["epochs"]=total_epoch
|
415 |
+
data["train"]["text_low_lr_rate"]=text_low_lr_rate
|
416 |
+
data["train"]["pretrained_s2G"]=pretrained_s2G
|
417 |
+
data["train"]["pretrained_s2D"]=pretrained_s2D
|
418 |
+
data["train"]["if_save_latest"]=if_save_latest
|
419 |
+
data["train"]["if_save_every_weights"]=if_save_every_weights
|
420 |
+
data["train"]["save_every_epoch"]=save_every_epoch
|
421 |
+
data["train"]["gpu_numbers"]=gpu_numbers1Ba
|
422 |
+
data["data"]["exp_dir"]=data["s2_ckpt_dir"]=s2_dir
|
423 |
+
data["save_weight_dir"]=SoVITS_weight_root
|
424 |
+
data["name"]=exp_name
|
425 |
+
tmp_config_path="%s/tmp_s2.json"%tmp
|
426 |
+
with open(tmp_config_path,"w")as f:f.write(json.dumps(data))
|
427 |
+
|
428 |
+
cmd = '"%s" GPT_SoVITS/s2_train.py --config "%s"'%(python_exec,tmp_config_path)
|
429 |
+
yield "SoVITS训练开始:%s"%cmd,{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
430 |
+
print(cmd)
|
431 |
+
p_train_SoVITS = Popen(cmd, shell=True)
|
432 |
+
p_train_SoVITS.wait()
|
433 |
+
p_train_SoVITS=None
|
434 |
+
yield "SoVITS训练完成",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
435 |
+
else:
|
436 |
+
yield "已有正在进行的SoVITS训练任务,需先终止才能开启下一次任务",{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
437 |
+
|
438 |
+
def close1Ba():
|
439 |
+
global p_train_SoVITS
|
440 |
+
if(p_train_SoVITS!=None):
|
441 |
+
kill_process(p_train_SoVITS.pid)
|
442 |
+
p_train_SoVITS=None
|
443 |
+
return "已终止SoVITS训练",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
444 |
+
|
445 |
+
p_train_GPT=None
|
446 |
+
def open1Bb(batch_size,total_epoch,exp_name,if_dpo,if_save_latest,if_save_every_weights,save_every_epoch,gpu_numbers,pretrained_s1):
|
447 |
+
global p_train_GPT
|
448 |
+
if(p_train_GPT==None):
|
449 |
+
with open("GPT_SoVITS/configs/s1longer.yaml")as f:
|
450 |
+
data=f.read()
|
451 |
+
data=yaml.load(data, Loader=yaml.FullLoader)
|
452 |
+
s1_dir="%s/%s"%(exp_root,exp_name)
|
453 |
+
os.makedirs("%s/logs_s1"%(s1_dir),exist_ok=True)
|
454 |
+
if(is_half==False):
|
455 |
+
data["train"]["precision"]="32"
|
456 |
+
batch_size = max(1, batch_size // 2)
|
457 |
+
data["train"]["batch_size"]=batch_size
|
458 |
+
data["train"]["epochs"]=total_epoch
|
459 |
+
data["pretrained_s1"]=pretrained_s1
|
460 |
+
data["train"]["save_every_n_epoch"]=save_every_epoch
|
461 |
+
data["train"]["if_save_every_weights"]=if_save_every_weights
|
462 |
+
data["train"]["if_save_latest"]=if_save_latest
|
463 |
+
data["train"]["if_dpo"]=if_dpo
|
464 |
+
data["train"]["half_weights_save_dir"]=GPT_weight_root
|
465 |
+
data["train"]["exp_name"]=exp_name
|
466 |
+
data["train_semantic_path"]="%s/6-name2semantic.tsv"%s1_dir
|
467 |
+
data["train_phoneme_path"]="%s/2-name2text.txt"%s1_dir
|
468 |
+
data["output_dir"]="%s/logs_s1"%s1_dir
|
469 |
+
|
470 |
+
os.environ["_CUDA_VISIBLE_DEVICES"]=gpu_numbers.replace("-",",")
|
471 |
+
os.environ["hz"]="25hz"
|
472 |
+
tmp_config_path="%s/tmp_s1.yaml"%tmp
|
473 |
+
with open(tmp_config_path, "w") as f:f.write(yaml.dump(data, default_flow_style=False))
|
474 |
+
# cmd = '"%s" GPT_SoVITS/s1_train.py --config_file "%s" --train_semantic_path "%s/6-name2semantic.tsv" --train_phoneme_path "%s/2-name2text.txt" --output_dir "%s/logs_s1"'%(python_exec,tmp_config_path,s1_dir,s1_dir,s1_dir)
|
475 |
+
cmd = '"%s" GPT_SoVITS/s1_train.py --config_file "%s" '%(python_exec,tmp_config_path)
|
476 |
+
yield "GPT训练开始:%s"%cmd,{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
477 |
+
print(cmd)
|
478 |
+
p_train_GPT = Popen(cmd, shell=True)
|
479 |
+
p_train_GPT.wait()
|
480 |
+
p_train_GPT=None
|
481 |
+
yield "GPT训练完成",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
482 |
+
else:
|
483 |
+
yield "已有正在进行的GPT训练任务,需先终止才能开启下一次任务",{"__type__":"update","visible":False},{"__type__":"update","visible":True}
|
484 |
+
|
485 |
+
def close1Bb():
|
486 |
+
global p_train_GPT
|
487 |
+
if(p_train_GPT!=None):
|
488 |
+
kill_process(p_train_GPT.pid)
|
489 |
+
p_train_GPT=None
|
490 |
+
return "已终止GPT训练",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
491 |
+
|
492 |
+
ps_slice=[]
|
493 |
+
def open_slice(inp,opt_root,threshold,min_length,min_interval,hop_size,max_sil_kept,_max,alpha,n_parts):
|
494 |
+
global ps_slice
|
495 |
+
inp = my_utils.clean_path(inp)
|
496 |
+
opt_root = my_utils.clean_path(opt_root)
|
497 |
+
if(os.path.exists(inp)==False):
|
498 |
+
yield "输入路径不存在",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
499 |
+
return
|
500 |
+
if os.path.isfile(inp):n_parts=1
|
501 |
+
elif os.path.isdir(inp):pass
|
502 |
+
else:
|
503 |
+
yield "输入路径存在但既不是文件也不是文件夹",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
504 |
+
return
|
505 |
+
if (ps_slice == []):
|
506 |
+
for i_part in range(n_parts):
|
507 |
+
cmd = '"%s" tools/slice_audio.py "%s" "%s" %s %s %s %s %s %s %s %s %s''' % (python_exec,inp, opt_root, threshold, min_length, min_interval, hop_size, max_sil_kept, _max, alpha, i_part, n_parts)
|
508 |
+
print(cmd)
|
509 |
+
p = Popen(cmd, shell=True)
|
510 |
+
ps_slice.append(p)
|
511 |
+
yield "切割执行中", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
512 |
+
for p in ps_slice:
|
513 |
+
p.wait()
|
514 |
+
ps_slice=[]
|
515 |
+
yield "切割结束",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
516 |
+
else:
|
517 |
+
yield "已有正在进行的切割任务,需先终止才能开启下一次任务", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
518 |
+
|
519 |
+
def close_slice():
|
520 |
+
global ps_slice
|
521 |
+
if (ps_slice != []):
|
522 |
+
for p_slice in ps_slice:
|
523 |
+
try:
|
524 |
+
kill_process(p_slice.pid)
|
525 |
+
except:
|
526 |
+
traceback.print_exc()
|
527 |
+
ps_slice=[]
|
528 |
+
return "已终止所有切割进程", {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
529 |
+
|
530 |
+
ps1a=[]
|
531 |
+
def open1a(inp_text,inp_wav_dir,exp_name,gpu_numbers,bert_pretrained_dir):
|
532 |
+
global ps1a
|
533 |
+
inp_text = my_utils.clean_path(inp_text)
|
534 |
+
inp_wav_dir = my_utils.clean_path(inp_wav_dir)
|
535 |
+
if (ps1a == []):
|
536 |
+
opt_dir="%s/%s"%(exp_root,exp_name)
|
537 |
+
config={
|
538 |
+
"inp_text":inp_text,
|
539 |
+
"inp_wav_dir":inp_wav_dir,
|
540 |
+
"exp_name":exp_name,
|
541 |
+
"opt_dir":opt_dir,
|
542 |
+
"bert_pretrained_dir":bert_pretrained_dir,
|
543 |
+
}
|
544 |
+
gpu_names=gpu_numbers.split("-")
|
545 |
+
all_parts=len(gpu_names)
|
546 |
+
for i_part in range(all_parts):
|
547 |
+
config.update(
|
548 |
+
{
|
549 |
+
"i_part": str(i_part),
|
550 |
+
"all_parts": str(all_parts),
|
551 |
+
"_CUDA_VISIBLE_DEVICES": gpu_names[i_part],
|
552 |
+
"is_half": str(is_half)
|
553 |
+
}
|
554 |
+
)
|
555 |
+
os.environ.update(config)
|
556 |
+
cmd = '"%s" GPT_SoVITS/prepare_datasets/1-get-text.py'%python_exec
|
557 |
+
print(cmd)
|
558 |
+
p = Popen(cmd, shell=True)
|
559 |
+
ps1a.append(p)
|
560 |
+
yield "文本进程执行中", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
561 |
+
for p in ps1a:
|
562 |
+
p.wait()
|
563 |
+
opt = []
|
564 |
+
for i_part in range(all_parts):
|
565 |
+
txt_path = "%s/2-name2text-%s.txt" % (opt_dir, i_part)
|
566 |
+
with open(txt_path, "r", encoding="utf8") as f:
|
567 |
+
opt += f.read().strip("\n").split("\n")
|
568 |
+
os.remove(txt_path)
|
569 |
+
path_text = "%s/2-name2text.txt" % opt_dir
|
570 |
+
with open(path_text, "w", encoding="utf8") as f:
|
571 |
+
f.write("\n".join(opt) + "\n")
|
572 |
+
ps1a=[]
|
573 |
+
yield "文本进程结束",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
574 |
+
else:
|
575 |
+
yield "已有正在进行的文本任务,需先终止才能开启下一次任务", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
576 |
+
|
577 |
+
def close1a():
|
578 |
+
global ps1a
|
579 |
+
if (ps1a != []):
|
580 |
+
for p1a in ps1a:
|
581 |
+
try:
|
582 |
+
kill_process(p1a.pid)
|
583 |
+
except:
|
584 |
+
traceback.print_exc()
|
585 |
+
ps1a=[]
|
586 |
+
return "已终止所有1a进程", {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
587 |
+
|
588 |
+
ps1b=[]
|
589 |
+
def open1b(inp_text,inp_wav_dir,exp_name,gpu_numbers,ssl_pretrained_dir):
|
590 |
+
global ps1b
|
591 |
+
inp_text = my_utils.clean_path(inp_text)
|
592 |
+
inp_wav_dir = my_utils.clean_path(inp_wav_dir)
|
593 |
+
if (ps1b == []):
|
594 |
+
config={
|
595 |
+
"inp_text":inp_text,
|
596 |
+
"inp_wav_dir":inp_wav_dir,
|
597 |
+
"exp_name":exp_name,
|
598 |
+
"opt_dir":"%s/%s"%(exp_root,exp_name),
|
599 |
+
"cnhubert_base_dir":ssl_pretrained_dir,
|
600 |
+
"is_half": str(is_half)
|
601 |
+
}
|
602 |
+
gpu_names=gpu_numbers.split("-")
|
603 |
+
all_parts=len(gpu_names)
|
604 |
+
for i_part in range(all_parts):
|
605 |
+
config.update(
|
606 |
+
{
|
607 |
+
"i_part": str(i_part),
|
608 |
+
"all_parts": str(all_parts),
|
609 |
+
"_CUDA_VISIBLE_DEVICES": gpu_names[i_part],
|
610 |
+
}
|
611 |
+
)
|
612 |
+
os.environ.update(config)
|
613 |
+
cmd = '"%s" GPT_SoVITS/prepare_datasets/2-get-hubert-wav32k.py'%python_exec
|
614 |
+
print(cmd)
|
615 |
+
p = Popen(cmd, shell=True)
|
616 |
+
ps1b.append(p)
|
617 |
+
yield "SSL提取进程执行中", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
618 |
+
for p in ps1b:
|
619 |
+
p.wait()
|
620 |
+
ps1b=[]
|
621 |
+
yield "SSL提取进程结束",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
622 |
+
else:
|
623 |
+
yield "已有正在进行的SSL提取任务,需先终止才能开启下一次任务", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
624 |
+
|
625 |
+
def close1b():
|
626 |
+
global ps1b
|
627 |
+
if (ps1b != []):
|
628 |
+
for p1b in ps1b:
|
629 |
+
try:
|
630 |
+
kill_process(p1b.pid)
|
631 |
+
except:
|
632 |
+
traceback.print_exc()
|
633 |
+
ps1b=[]
|
634 |
+
return "已终止所有1b进程", {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
635 |
+
|
636 |
+
ps1c=[]
|
637 |
+
def open1c(inp_text,exp_name,gpu_numbers,pretrained_s2G_path):
|
638 |
+
global ps1c
|
639 |
+
inp_text = my_utils.clean_path(inp_text)
|
640 |
+
if (ps1c == []):
|
641 |
+
opt_dir="%s/%s"%(exp_root,exp_name)
|
642 |
+
config={
|
643 |
+
"inp_text":inp_text,
|
644 |
+
"exp_name":exp_name,
|
645 |
+
"opt_dir":opt_dir,
|
646 |
+
"pretrained_s2G":pretrained_s2G_path,
|
647 |
+
"s2config_path":"GPT_SoVITS/configs/s2.json",
|
648 |
+
"is_half": str(is_half)
|
649 |
+
}
|
650 |
+
gpu_names=gpu_numbers.split("-")
|
651 |
+
all_parts=len(gpu_names)
|
652 |
+
for i_part in range(all_parts):
|
653 |
+
config.update(
|
654 |
+
{
|
655 |
+
"i_part": str(i_part),
|
656 |
+
"all_parts": str(all_parts),
|
657 |
+
"_CUDA_VISIBLE_DEVICES": gpu_names[i_part],
|
658 |
+
}
|
659 |
+
)
|
660 |
+
os.environ.update(config)
|
661 |
+
cmd = '"%s" GPT_SoVITS/prepare_datasets/3-get-semantic.py'%python_exec
|
662 |
+
print(cmd)
|
663 |
+
p = Popen(cmd, shell=True)
|
664 |
+
ps1c.append(p)
|
665 |
+
yield "语义token提取进程执行中", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
666 |
+
for p in ps1c:
|
667 |
+
p.wait()
|
668 |
+
opt = ["item_name\tsemantic_audio"]
|
669 |
+
path_semantic = "%s/6-name2semantic.tsv" % opt_dir
|
670 |
+
for i_part in range(all_parts):
|
671 |
+
semantic_path = "%s/6-name2semantic-%s.tsv" % (opt_dir, i_part)
|
672 |
+
with open(semantic_path, "r", encoding="utf8") as f:
|
673 |
+
opt += f.read().strip("\n").split("\n")
|
674 |
+
os.remove(semantic_path)
|
675 |
+
with open(path_semantic, "w", encoding="utf8") as f:
|
676 |
+
f.write("\n".join(opt) + "\n")
|
677 |
+
ps1c=[]
|
678 |
+
yield "语义token提取进程结束",{"__type__":"update","visible":True},{"__type__":"update","visible":False}
|
679 |
+
else:
|
680 |
+
yield "已有正在进行的语义token提取任务,需先终止才能开启下一次任务", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
681 |
+
|
682 |
+
def close1c():
|
683 |
+
global ps1c
|
684 |
+
if (ps1c != []):
|
685 |
+
for p1c in ps1c:
|
686 |
+
try:
|
687 |
+
kill_process(p1c.pid)
|
688 |
+
except:
|
689 |
+
traceback.print_exc()
|
690 |
+
ps1c=[]
|
691 |
+
return "已终止所有语义token进程", {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
692 |
+
#####inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numbers1c,bert_pretrained_dir,cnhubert_base_dir,pretrained_s2G
|
693 |
+
ps1abc=[]
|
694 |
+
def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numbers1c,bert_pretrained_dir,ssl_pretrained_dir,pretrained_s2G_path):
|
695 |
+
global ps1abc
|
696 |
+
inp_text = my_utils.clean_path(inp_text)
|
697 |
+
inp_wav_dir = my_utils.clean_path(inp_wav_dir)
|
698 |
+
if (ps1abc == []):
|
699 |
+
opt_dir="%s/%s"%(exp_root,exp_name)
|
700 |
+
try:
|
701 |
+
#############################1a
|
702 |
+
path_text="%s/2-name2text.txt" % opt_dir
|
703 |
+
if(os.path.exists(path_text)==False or (os.path.exists(path_text)==True and len(open(path_text,"r",encoding="utf8").read().strip("\n").split("\n"))<2)):
|
704 |
+
config={
|
705 |
+
"inp_text":inp_text,
|
706 |
+
"inp_wav_dir":inp_wav_dir,
|
707 |
+
"exp_name":exp_name,
|
708 |
+
"opt_dir":opt_dir,
|
709 |
+
"bert_pretrained_dir":bert_pretrained_dir,
|
710 |
+
"is_half": str(is_half)
|
711 |
+
}
|
712 |
+
gpu_names=gpu_numbers1a.split("-")
|
713 |
+
all_parts=len(gpu_names)
|
714 |
+
for i_part in range(all_parts):
|
715 |
+
config.update(
|
716 |
+
{
|
717 |
+
"i_part": str(i_part),
|
718 |
+
"all_parts": str(all_parts),
|
719 |
+
"_CUDA_VISIBLE_DEVICES": gpu_names[i_part],
|
720 |
+
}
|
721 |
+
)
|
722 |
+
os.environ.update(config)
|
723 |
+
cmd = '"%s" GPT_SoVITS/prepare_datasets/1-get-text.py'%python_exec
|
724 |
+
print(cmd)
|
725 |
+
p = Popen(cmd, shell=True)
|
726 |
+
ps1abc.append(p)
|
727 |
+
yield "进度:1a-ing", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
728 |
+
for p in ps1abc:p.wait()
|
729 |
+
|
730 |
+
opt = []
|
731 |
+
for i_part in range(all_parts):#txt_path="%s/2-name2text-%s.txt"%(opt_dir,i_part)
|
732 |
+
txt_path = "%s/2-name2text-%s.txt" % (opt_dir, i_part)
|
733 |
+
with open(txt_path, "r",encoding="utf8") as f:
|
734 |
+
opt += f.read().strip("\n").split("\n")
|
735 |
+
os.remove(txt_path)
|
736 |
+
with open(path_text, "w",encoding="utf8") as f:
|
737 |
+
f.write("\n".join(opt) + "\n")
|
738 |
+
|
739 |
+
yield "进度:1a-done", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
740 |
+
ps1abc=[]
|
741 |
+
#############################1b
|
742 |
+
config={
|
743 |
+
"inp_text":inp_text,
|
744 |
+
"inp_wav_dir":inp_wav_dir,
|
745 |
+
"exp_name":exp_name,
|
746 |
+
"opt_dir":opt_dir,
|
747 |
+
"cnhubert_base_dir":ssl_pretrained_dir,
|
748 |
+
}
|
749 |
+
gpu_names=gpu_numbers1Ba.split("-")
|
750 |
+
all_parts=len(gpu_names)
|
751 |
+
for i_part in range(all_parts):
|
752 |
+
config.update(
|
753 |
+
{
|
754 |
+
"i_part": str(i_part),
|
755 |
+
"all_parts": str(all_parts),
|
756 |
+
"_CUDA_VISIBLE_DEVICES": gpu_names[i_part],
|
757 |
+
}
|
758 |
+
)
|
759 |
+
os.environ.update(config)
|
760 |
+
cmd = '"%s" GPT_SoVITS/prepare_datasets/2-get-hubert-wav32k.py'%python_exec
|
761 |
+
print(cmd)
|
762 |
+
p = Popen(cmd, shell=True)
|
763 |
+
ps1abc.append(p)
|
764 |
+
yield "进度:1a-done, 1b-ing", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
765 |
+
for p in ps1abc:p.wait()
|
766 |
+
yield "进度:1a1b-done", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
767 |
+
ps1abc=[]
|
768 |
+
#############################1c
|
769 |
+
path_semantic = "%s/6-name2semantic.tsv" % opt_dir
|
770 |
+
if(os.path.exists(path_semantic)==False or (os.path.exists(path_semantic)==True and os.path.getsize(path_semantic)<31)):
|
771 |
+
config={
|
772 |
+
"inp_text":inp_text,
|
773 |
+
"exp_name":exp_name,
|
774 |
+
"opt_dir":opt_dir,
|
775 |
+
"pretrained_s2G":pretrained_s2G_path,
|
776 |
+
"s2config_path":"GPT_SoVITS/configs/s2.json",
|
777 |
+
}
|
778 |
+
gpu_names=gpu_numbers1c.split("-")
|
779 |
+
all_parts=len(gpu_names)
|
780 |
+
for i_part in range(all_parts):
|
781 |
+
config.update(
|
782 |
+
{
|
783 |
+
"i_part": str(i_part),
|
784 |
+
"all_parts": str(all_parts),
|
785 |
+
"_CUDA_VISIBLE_DEVICES": gpu_names[i_part],
|
786 |
+
}
|
787 |
+
)
|
788 |
+
os.environ.update(config)
|
789 |
+
cmd = '"%s" GPT_SoVITS/prepare_datasets/3-get-semantic.py'%python_exec
|
790 |
+
print(cmd)
|
791 |
+
p = Popen(cmd, shell=True)
|
792 |
+
ps1abc.append(p)
|
793 |
+
yield "进度:1a1b-done, 1cing", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
794 |
+
for p in ps1abc:p.wait()
|
795 |
+
|
796 |
+
opt = ["item_name\tsemantic_audio"]
|
797 |
+
for i_part in range(all_parts):
|
798 |
+
semantic_path = "%s/6-name2semantic-%s.tsv" % (opt_dir, i_part)
|
799 |
+
with open(semantic_path, "r",encoding="utf8") as f:
|
800 |
+
opt += f.read().strip("\n").split("\n")
|
801 |
+
os.remove(semantic_path)
|
802 |
+
with open(path_semantic, "w",encoding="utf8") as f:
|
803 |
+
f.write("\n".join(opt) + "\n")
|
804 |
+
yield "进度:all-done", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
805 |
+
ps1abc = []
|
806 |
+
yield "一键三连进程结束", {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
807 |
+
except:
|
808 |
+
traceback.print_exc()
|
809 |
+
close1abc()
|
810 |
+
yield "一键三连中途报错", {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
811 |
+
else:
|
812 |
+
yield "已有正在进行的一键三连任务,需先终止才能开启下一次任务", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
813 |
+
|
814 |
+
def close1abc():
|
815 |
+
global ps1abc
|
816 |
+
if (ps1abc != []):
|
817 |
+
for p1abc in ps1abc:
|
818 |
+
try:
|
819 |
+
kill_process(p1abc.pid)
|
820 |
+
except:
|
821 |
+
traceback.print_exc()
|
822 |
+
ps1abc=[]
|
823 |
+
return "已终止所有一键三连进程", {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
824 |
+
|
825 |
+
with gr.Blocks(title="GPT-SoVITS WebUI") as app:
|
826 |
+
gr.Markdown("# <center>🌊💕🎶 XTTS 微调:2分钟语音,开启中日英16种语言真实拟声</center>")
|
827 |
+
gr.Markdown("## <center>🌟 只需2分钟的语音,一键在线微调 最强多语种模型</center>")
|
828 |
+
gr.Markdown("### <center>🤗 更多精彩,尽在[滔滔AI](https://www.talktalkai.com/);滔滔AI,��爱滔滔!💕</center>")
|
829 |
+
|
830 |
+
with gr.Tabs():
|
831 |
+
with gr.TabItem(i18n("1 - 制作数据集")):#提前随机切片防止uvr5爆内存->uvr5->slicer->asr->打标
|
832 |
+
#gr.Markdown(value=i18n("0a-UVR5人声伴奏分离&去混响去延迟工具"))
|
833 |
+
with gr.Row():
|
834 |
+
if_uvr5 = gr.Checkbox(label=i18n("是否开启UVR5-WebUI"),show_label=True, visible=False)
|
835 |
+
uvr5_info = gr.Textbox(label=i18n("UVR5进程输出信息"), visible=False)
|
836 |
+
gr.Markdown(value=i18n("1a-语音切分工具"))
|
837 |
+
with gr.Row():
|
838 |
+
with gr.Row():
|
839 |
+
slice_inp_path=gr.Textbox(label=i18n("音频自动切分输入路径,可文件可文件夹"),info="您需要先在xtts-v2文件夹中上传训练音频,如jay.wav;音频时长建议大于2分钟",value="",placeholder="jay.wav")
|
840 |
+
slice_opt_root=gr.Textbox(label=i18n("切分后的子音频的输出根目录"),value="output/slicer_opt")
|
841 |
+
threshold=gr.Textbox(label=i18n("threshold:音量小于这个值视作静音的备选切割点"),value="-34")
|
842 |
+
min_length=gr.Textbox(label=i18n("min_length:每段最小多长,如果第一段太短一直和后面段连起来直到超过这个值"),value="4000")
|
843 |
+
min_interval=gr.Textbox(label=i18n("min_interval:最短切割间隔"),value="300")
|
844 |
+
hop_size=gr.Textbox(label=i18n("hop_size:怎么算音量曲线,越小精度越大计算量越高(不是精度越大效果越好)"),value="10")
|
845 |
+
max_sil_kept=gr.Textbox(label=i18n("max_sil_kept:切完后静音最多留多长"),value="500")
|
846 |
+
with gr.Row():
|
847 |
+
open_slicer_button=gr.Button(i18n("1. 开启语音切割"), variant="primary",visible=True)
|
848 |
+
close_slicer_button=gr.Button(i18n("终止语音切割"), variant="primary",visible=False)
|
849 |
+
_max=gr.Slider(minimum=0,maximum=1,step=0.05,label=i18n("max:归一化后最大值多少"),value=0.9,interactive=True)
|
850 |
+
alpha=gr.Slider(minimum=0,maximum=1,step=0.05,label=i18n("alpha_mix:混多少比例归一化后音频进来"),value=0.25,interactive=True)
|
851 |
+
n_process=gr.Slider(minimum=1,maximum=n_cpu,step=1,label=i18n("切割使用的进程数"),value=4,interactive=True)
|
852 |
+
slicer_info = gr.Textbox(label=i18n("语音切割进程输出信息"))
|
853 |
+
#gr.Markdown(value=i18n("0bb-语音降噪工具"))
|
854 |
+
with gr.Row():
|
855 |
+
open_denoise_button = gr.Button(i18n("开启语音降噪"), visible=False)
|
856 |
+
close_denoise_button = gr.Button(i18n("终止语音降噪进程"), variant="primary",visible=False)
|
857 |
+
denoise_input_dir=gr.Textbox(label=i18n("降噪音频文件输入文件夹"),value="", visible=False)
|
858 |
+
denoise_output_dir=gr.Textbox(label=i18n("降噪结果输出文件夹"),value="output/denoise_opt", visible=False)
|
859 |
+
denoise_info = gr.Textbox(label=i18n("语音降噪进程输出信息"), visible=False)
|
860 |
+
gr.Markdown(value=i18n("1b-批量语音识别"))
|
861 |
+
with gr.Row():
|
862 |
+
open_asr_button = gr.Button(i18n("2. 开启离线批量ASR"), variant="primary",visible=True)
|
863 |
+
close_asr_button = gr.Button(i18n("终止ASR进程"), variant="primary",visible=False)
|
864 |
+
with gr.Column():
|
865 |
+
with gr.Row():
|
866 |
+
asr_inp_dir = gr.Textbox(
|
867 |
+
label=i18n("输入文件夹路径"),
|
868 |
+
value="output/slicer_opt",
|
869 |
+
interactive=True,
|
870 |
+
)
|
871 |
+
asr_opt_dir = gr.Textbox(
|
872 |
+
label = i18n("输出文件夹路径"),
|
873 |
+
value = "output/asr_opt",
|
874 |
+
interactive = True,
|
875 |
+
)
|
876 |
+
with gr.Row():
|
877 |
+
asr_model = gr.Dropdown(
|
878 |
+
label = i18n("ASR 模型"),
|
879 |
+
choices = list(asr_dict.keys()),
|
880 |
+
interactive = True,
|
881 |
+
value="达摩 ASR (中文)"
|
882 |
+
)
|
883 |
+
asr_size = gr.Dropdown(
|
884 |
+
label = i18n("ASR 模型尺寸"),
|
885 |
+
choices = ["large"],
|
886 |
+
interactive = True,
|
887 |
+
value="large"
|
888 |
+
)
|
889 |
+
asr_lang = gr.Dropdown(
|
890 |
+
label = i18n("ASR 语言设置"),
|
891 |
+
choices = ["zh"],
|
892 |
+
interactive = True,
|
893 |
+
value="zh"
|
894 |
+
)
|
895 |
+
lang = asr_lang
|
896 |
+
with gr.Row():
|
897 |
+
asr_info = gr.Textbox(label=i18n("ASR进程输出信息"))
|
898 |
+
|
899 |
+
def change_lang_choices(key): #根据选择的模型修改可选的语言
|
900 |
+
# return gr.Dropdown(choices=asr_dict[key]['lang'])
|
901 |
+
return {"__type__": "update", "choices": asr_dict[key]['lang'],"value":asr_dict[key]['lang'][0]}
|
902 |
+
def change_size_choices(key): # 根据选择的模型修改可选的模型尺寸
|
903 |
+
# return gr.Dropdown(choices=asr_dict[key]['size'])
|
904 |
+
return {"__type__": "update", "choices": asr_dict[key]['size']}
|
905 |
+
asr_model.change(change_lang_choices, [asr_model], [asr_lang])
|
906 |
+
asr_model.change(change_size_choices, [asr_model], [asr_size])
|
907 |
+
|
908 |
+
gr.Markdown(value=i18n("1c-语音文本校对标注工具"))
|
909 |
+
with gr.Row():
|
910 |
+
if_label = gr.Checkbox(label=i18n("是否开启打标WebUI"),show_label=True)
|
911 |
+
path_list = gr.Textbox(
|
912 |
+
label=i18n(".list标注文件的路径"),
|
913 |
+
value="output/asr_opt/slicer_opt.list",
|
914 |
+
interactive=True,
|
915 |
+
)
|
916 |
+
label_info = gr.Textbox(label=i18n("打标工具进程输出信息"))
|
917 |
+
if_label.change(change_label, [if_label,path_list], [label_info])
|
918 |
+
if_uvr5.change(change_uvr5, [if_uvr5], [uvr5_info])
|
919 |
+
open_asr_button.click(open_asr, [asr_inp_dir, asr_opt_dir, asr_model, asr_size, asr_lang], [asr_info,open_asr_button,close_asr_button])
|
920 |
+
close_asr_button.click(close_asr, [], [asr_info,open_asr_button,close_asr_button])
|
921 |
+
open_slicer_button.click(open_slice, [slice_inp_path,slice_opt_root,threshold,min_length,min_interval,hop_size,max_sil_kept,_max,alpha,n_process], [slicer_info,open_slicer_button,close_slicer_button])
|
922 |
+
close_slicer_button.click(close_slice, [], [slicer_info,open_slicer_button,close_slicer_button])
|
923 |
+
open_denoise_button.click(open_denoise, [denoise_input_dir,denoise_output_dir], [denoise_info,open_denoise_button,close_denoise_button])
|
924 |
+
close_denoise_button.click(close_denoise, [], [denoise_info,open_denoise_button,close_denoise_button])
|
925 |
+
|
926 |
+
with gr.Tab("2 - XTTS模型微调"):
|
927 |
+
inp_list_path_value = str(Path.cwd() / "output/asr_opt/slicer_opt.list")
|
928 |
+
out_csv_path_value = str(Path.cwd() / "output.csv")
|
929 |
+
inp_list_path = gr.Textbox(value=inp_list_path_value, label=".list文件地址")
|
930 |
+
out_csv_path = gr.Textbox(value=out_csv_path_value, label=".csv文件地址")
|
931 |
+
list_to_csv = gr.Button("3. 准备训练csv文件", variant="primary")
|
932 |
+
train_csv = gr.Textbox(
|
933 |
+
label="训练数据集csv文件",
|
934 |
+
)
|
935 |
+
eval_csv = gr.Textbox(
|
936 |
+
label="评价数据集csv文件",
|
937 |
+
)
|
938 |
+
list_to_csv.click(convert_list_to_csv, [inp_list_path, out_csv_path], [train_csv, eval_csv])
|
939 |
+
out_path_value = str(Path.cwd() / "finetune_models")
|
940 |
+
out_path = gr.Textbox(value=out_path_value, label="XTTS微调模型的文件夹")
|
941 |
+
num_epochs = gr.Slider(
|
942 |
+
label="训练步数 Number of epochs:",
|
943 |
+
minimum=1,
|
944 |
+
maximum=100,
|
945 |
+
step=1,
|
946 |
+
value=6,
|
947 |
+
)
|
948 |
+
batch_size = gr.Slider(
|
949 |
+
label="Batch size:",
|
950 |
+
minimum=2,
|
951 |
+
maximum=512,
|
952 |
+
step=1,
|
953 |
+
value=2,
|
954 |
+
)
|
955 |
+
grad_acumm = gr.Slider(
|
956 |
+
label="Grad accumulation steps:",
|
957 |
+
minimum=1,
|
958 |
+
maximum=128,
|
959 |
+
step=1,
|
960 |
+
value=1,
|
961 |
+
)
|
962 |
+
max_audio_length = gr.Slider(
|
963 |
+
label="Max permitted audio size in seconds:",
|
964 |
+
minimum=2,
|
965 |
+
maximum=20,
|
966 |
+
step=1,
|
967 |
+
value=11,
|
968 |
+
visible=False,
|
969 |
+
)
|
970 |
+
progress_train = gr.Label(
|
971 |
+
label="训练进程"
|
972 |
+
)
|
973 |
+
logs_tts_train = gr.Textbox(
|
974 |
+
label="训练详细信息",
|
975 |
+
interactive=False,
|
976 |
+
)
|
977 |
+
app.load(read_logs, None, logs_tts_train, every=1)
|
978 |
+
train_btn = gr.Button(value="4. 开始模型训练", variant="primary")
|
979 |
+
|
980 |
+
def train_model(language, train_csv, eval_csv, num_epochs, batch_size, grad_acumm, output_path, max_audio_length):
|
981 |
+
print(f"开始训练,训练素材的语种为:{language}")
|
982 |
+
clear_gpu_cache()
|
983 |
+
if not train_csv or not eval_csv:
|
984 |
+
return "You need to run the data processing step or manually set `Train CSV` and `Eval CSV` fields !", "", "", "", ""
|
985 |
+
try:
|
986 |
+
# convert seconds to waveform frames
|
987 |
+
max_audio_length = int(max_audio_length * 22050)
|
988 |
+
config_path, original_xtts_checkpoint, vocab_file, exp_path, speaker_wav = train_gpt(language, num_epochs, batch_size, grad_acumm, train_csv, eval_csv, output_path=output_path, max_audio_length=max_audio_length)
|
989 |
+
except:
|
990 |
+
traceback.print_exc()
|
991 |
+
error = traceback.format_exc()
|
992 |
+
return f"The training was interrupted due an error !! Please check the console to check the full error message! \n Error summary: {error}", "", "", "", ""
|
993 |
+
|
994 |
+
# copy original files to avoid parameters changes issues
|
995 |
+
os.system(f"cp {config_path} {exp_path}")
|
996 |
+
os.system(f"cp {vocab_file} {exp_path}")
|
997 |
+
|
998 |
+
ft_xtts_checkpoint = os.path.join(exp_path, "best_model.pth")
|
999 |
+
print("模型已成功微调!")
|
1000 |
+
clear_gpu_cache()
|
1001 |
+
ref_audio_names = os.listdir("kobe/Kobe")
|
1002 |
+
ref_audio_list = [os.path.join("kobe/Kobe", ref_audio_name) for ref_audio_name in ref_audio_names]
|
1003 |
+
first_five_ref_audio = "\n".join(ref_audio_list[0:8])
|
1004 |
+
return "模型已成功微调!", config_path, vocab_file, ft_xtts_checkpoint, first_five_ref_audio, speaker_wav
|
1005 |
+
|
1006 |
+
with gr.Tab("3 - XTTS语音合成"):
|
1007 |
+
with gr.Row():
|
1008 |
+
with gr.Column() as col1:
|
1009 |
+
xtts_checkpoint = gr.Textbox(
|
1010 |
+
label="XTTS checkpoint 路径",
|
1011 |
+
value="",
|
1012 |
+
)
|
1013 |
+
xtts_config = gr.Textbox(
|
1014 |
+
label="XTTS config 路径",
|
1015 |
+
value="",
|
1016 |
+
)
|
1017 |
+
|
1018 |
+
xtts_vocab = gr.Textbox(
|
1019 |
+
label="XTTS vocab 路径",
|
1020 |
+
value="",
|
1021 |
+
)
|
1022 |
+
progress_load = gr.Label(
|
1023 |
+
label="模型加载进程"
|
1024 |
+
)
|
1025 |
+
load_btn = gr.Button(value="5. 加载已训练好的模型", variant="primary")
|
1026 |
+
|
1027 |
+
with gr.Column() as col2:
|
1028 |
+
first_five_speaker_reference_audio = gr.Textbox(label="您可以选用的参考音频", visible=True, interactive=True)
|
1029 |
+
speaker_reference_audio = gr.Textbox(
|
1030 |
+
label="您正在使用的参考音频",
|
1031 |
+
info="不同参考音频对应的合成效果不同。您可以尝试多次,每次填写一条音频路径",
|
1032 |
+
value="",
|
1033 |
+
)
|
1034 |
+
tts_text = gr.Textbox(
|
1035 |
+
label="请填写语音合成的文本🍻",
|
1036 |
+
placeholder="想说却还没说的,还很多",
|
1037 |
+
)
|
1038 |
+
tts_language = gr.Dropdown(
|
1039 |
+
label="请选择文本对应的语言",
|
1040 |
+
value="zh",
|
1041 |
+
choices=[
|
1042 |
+
"en",
|
1043 |
+
"es",
|
1044 |
+
"fr",
|
1045 |
+
"de",
|
1046 |
+
"it",
|
1047 |
+
"pt",
|
1048 |
+
"pl",
|
1049 |
+
"tr",
|
1050 |
+
"ru",
|
1051 |
+
"nl",
|
1052 |
+
"cs",
|
1053 |
+
"ar",
|
1054 |
+
"zh",
|
1055 |
+
"hu",
|
1056 |
+
"ko",
|
1057 |
+
"ja",
|
1058 |
+
]
|
1059 |
+
)
|
1060 |
+
|
1061 |
+
tts_btn = gr.Button(value="6. 开启AI语音之旅吧💕", variant="primary")
|
1062 |
+
|
1063 |
+
with gr.Column() as col3:
|
1064 |
+
progress_gen = gr.Label(
|
1065 |
+
label="语音合成进程"
|
1066 |
+
)
|
1067 |
+
tts_output_audio = gr.Audio(label="为您合成的专属音频🎶")
|
1068 |
+
reference_audio = gr.Audio(label="您使用的参考音频")
|
1069 |
+
|
1070 |
+
train_btn.click(
|
1071 |
+
fn=train_model,
|
1072 |
+
inputs=[
|
1073 |
+
lang,
|
1074 |
+
train_csv,
|
1075 |
+
eval_csv,
|
1076 |
+
num_epochs,
|
1077 |
+
batch_size,
|
1078 |
+
grad_acumm,
|
1079 |
+
out_path,
|
1080 |
+
max_audio_length,
|
1081 |
+
],
|
1082 |
+
outputs=[progress_train, xtts_config, xtts_vocab, xtts_checkpoint, first_five_speaker_reference_audio, speaker_reference_audio],
|
1083 |
+
)
|
1084 |
+
|
1085 |
+
load_btn.click(
|
1086 |
+
fn=load_model,
|
1087 |
+
inputs=[
|
1088 |
+
xtts_checkpoint,
|
1089 |
+
xtts_config,
|
1090 |
+
xtts_vocab
|
1091 |
+
],
|
1092 |
+
outputs=[progress_load],
|
1093 |
+
)
|
1094 |
+
|
1095 |
+
tts_btn.click(
|
1096 |
+
fn=run_tts,
|
1097 |
+
inputs=[
|
1098 |
+
tts_language,
|
1099 |
+
tts_text,
|
1100 |
+
speaker_reference_audio,
|
1101 |
+
],
|
1102 |
+
outputs=[progress_gen, tts_output_audio, reference_audio],
|
1103 |
+
)
|
1104 |
+
|
1105 |
+
gr.Markdown("### <center>注意❗:请不要生成会对个人以及组织造成侵��的内容,此程序仅供科研、学习及个人娱乐使用。请自觉合规使用此程序,程序开发者不负有任何责任。</center>")
|
1106 |
+
gr.HTML('''
|
1107 |
+
<div class="footer">
|
1108 |
+
<p>🌊🏞️🎶 - 江水东流急,滔滔无尽声。 明·顾璘
|
1109 |
+
</p>
|
1110 |
+
</div>
|
1111 |
+
''')
|
1112 |
+
app.queue().launch(
|
1113 |
+
share=True,
|
1114 |
+
show_error=True,
|
1115 |
+
)
|
finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/config.json
ADDED
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"output_path": "output",
|
3 |
+
"logger_uri": null,
|
4 |
+
"run_name": "run",
|
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 |
+
"save_on_interrupt": true,
|
13 |
+
"log_model_step": null,
|
14 |
+
"save_step": 10000,
|
15 |
+
"save_n_checkpoints": 5,
|
16 |
+
"save_checkpoints": true,
|
17 |
+
"save_all_best": false,
|
18 |
+
"save_best_after": 10000,
|
19 |
+
"target_loss": null,
|
20 |
+
"print_eval": false,
|
21 |
+
"test_delay_epochs": 0,
|
22 |
+
"run_eval": true,
|
23 |
+
"run_eval_steps": null,
|
24 |
+
"distributed_backend": "nccl",
|
25 |
+
"distributed_url": "tcp://localhost:54321",
|
26 |
+
"mixed_precision": false,
|
27 |
+
"precision": "fp16",
|
28 |
+
"epochs": 1000,
|
29 |
+
"batch_size": 32,
|
30 |
+
"eval_batch_size": 16,
|
31 |
+
"grad_clip": 0.0,
|
32 |
+
"scheduler_after_epoch": true,
|
33 |
+
"lr": 0.001,
|
34 |
+
"optimizer": "radam",
|
35 |
+
"optimizer_params": null,
|
36 |
+
"lr_scheduler": null,
|
37 |
+
"lr_scheduler_params": {},
|
38 |
+
"use_grad_scaler": false,
|
39 |
+
"allow_tf32": false,
|
40 |
+
"cudnn_enable": true,
|
41 |
+
"cudnn_deterministic": false,
|
42 |
+
"cudnn_benchmark": false,
|
43 |
+
"training_seed": 54321,
|
44 |
+
"model": "xtts",
|
45 |
+
"num_loader_workers": 0,
|
46 |
+
"num_eval_loader_workers": 0,
|
47 |
+
"use_noise_augment": false,
|
48 |
+
"audio": {
|
49 |
+
"sample_rate": 22050,
|
50 |
+
"output_sample_rate": 24000
|
51 |
+
},
|
52 |
+
"use_phonemes": false,
|
53 |
+
"phonemizer": null,
|
54 |
+
"phoneme_language": null,
|
55 |
+
"compute_input_seq_cache": false,
|
56 |
+
"text_cleaner": null,
|
57 |
+
"enable_eos_bos_chars": false,
|
58 |
+
"test_sentences_file": "",
|
59 |
+
"phoneme_cache_path": null,
|
60 |
+
"characters": null,
|
61 |
+
"add_blank": false,
|
62 |
+
"batch_group_size": 0,
|
63 |
+
"loss_masking": null,
|
64 |
+
"min_audio_len": 1,
|
65 |
+
"max_audio_len": Infinity,
|
66 |
+
"min_text_len": 1,
|
67 |
+
"max_text_len": Infinity,
|
68 |
+
"compute_f0": false,
|
69 |
+
"compute_energy": false,
|
70 |
+
"compute_linear_spec": false,
|
71 |
+
"precompute_num_workers": 0,
|
72 |
+
"start_by_longest": false,
|
73 |
+
"shuffle": false,
|
74 |
+
"drop_last": false,
|
75 |
+
"datasets": [
|
76 |
+
{
|
77 |
+
"formatter": "",
|
78 |
+
"dataset_name": "",
|
79 |
+
"path": "",
|
80 |
+
"meta_file_train": "",
|
81 |
+
"ignored_speakers": null,
|
82 |
+
"language": "",
|
83 |
+
"phonemizer": "",
|
84 |
+
"meta_file_val": "",
|
85 |
+
"meta_file_attn_mask": ""
|
86 |
+
}
|
87 |
+
],
|
88 |
+
"test_sentences": [],
|
89 |
+
"eval_split_max_size": null,
|
90 |
+
"eval_split_size": 0.01,
|
91 |
+
"use_speaker_weighted_sampler": false,
|
92 |
+
"speaker_weighted_sampler_alpha": 1.0,
|
93 |
+
"use_language_weighted_sampler": false,
|
94 |
+
"language_weighted_sampler_alpha": 1.0,
|
95 |
+
"use_length_weighted_sampler": false,
|
96 |
+
"length_weighted_sampler_alpha": 1.0,
|
97 |
+
"model_args": {
|
98 |
+
"gpt_batch_size": 1,
|
99 |
+
"enable_redaction": false,
|
100 |
+
"kv_cache": true,
|
101 |
+
"gpt_checkpoint": null,
|
102 |
+
"clvp_checkpoint": null,
|
103 |
+
"decoder_checkpoint": null,
|
104 |
+
"num_chars": 255,
|
105 |
+
"tokenizer_file": "",
|
106 |
+
"gpt_max_audio_tokens": 605,
|
107 |
+
"gpt_max_text_tokens": 402,
|
108 |
+
"gpt_max_prompt_tokens": 70,
|
109 |
+
"gpt_layers": 30,
|
110 |
+
"gpt_n_model_channels": 1024,
|
111 |
+
"gpt_n_heads": 16,
|
112 |
+
"gpt_number_text_tokens": 6681,
|
113 |
+
"gpt_start_text_token": null,
|
114 |
+
"gpt_stop_text_token": null,
|
115 |
+
"gpt_num_audio_tokens": 1026,
|
116 |
+
"gpt_start_audio_token": 1024,
|
117 |
+
"gpt_stop_audio_token": 1025,
|
118 |
+
"gpt_code_stride_len": 1024,
|
119 |
+
"gpt_use_masking_gt_prompt_approach": true,
|
120 |
+
"gpt_use_perceiver_resampler": true,
|
121 |
+
"input_sample_rate": 22050,
|
122 |
+
"output_sample_rate": 24000,
|
123 |
+
"output_hop_length": 256,
|
124 |
+
"decoder_input_dim": 1024,
|
125 |
+
"d_vector_dim": 512,
|
126 |
+
"cond_d_vector_in_each_upsampling_layer": true,
|
127 |
+
"duration_const": 102400
|
128 |
+
},
|
129 |
+
"model_dir": null,
|
130 |
+
"languages": [
|
131 |
+
"en",
|
132 |
+
"es",
|
133 |
+
"fr",
|
134 |
+
"de",
|
135 |
+
"it",
|
136 |
+
"pt",
|
137 |
+
"pl",
|
138 |
+
"tr",
|
139 |
+
"ru",
|
140 |
+
"nl",
|
141 |
+
"cs",
|
142 |
+
"ar",
|
143 |
+
"zh-cn",
|
144 |
+
"hu",
|
145 |
+
"ko",
|
146 |
+
"ja",
|
147 |
+
"hi"
|
148 |
+
],
|
149 |
+
"temperature": 0.75,
|
150 |
+
"length_penalty": 1.0,
|
151 |
+
"repetition_penalty": 5.0,
|
152 |
+
"top_k": 50,
|
153 |
+
"top_p": 0.85,
|
154 |
+
"num_gpt_outputs": 1,
|
155 |
+
"gpt_cond_len": 30,
|
156 |
+
"gpt_cond_chunk_len": 4,
|
157 |
+
"max_ref_len": 30,
|
158 |
+
"sound_norm_refs": false
|
159 |
+
}
|
finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/dvae.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b29bc227d410d4991e0a8c09b858f77415013eeb9fba9650258e96095557d97a
|
3 |
+
size 210514388
|
finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/mel_stats.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1f69422a8a8f344c4fca2f0c6b8d41d2151d6615b7321e48e6bb15ae949b119c
|
3 |
+
size 1067
|
finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/model.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c7ea20001c6a0a841c77e252d8409f6a74fb423e79b3206a0771ba5989776187
|
3 |
+
size 1867929118
|
finetune_models_kobe/run/training/XTTS_v2.0_original_model_files/vocab.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|