gnumanth commited on
Commit
9bb3a8f
·
verified ·
1 Parent(s): 5f7e4cf

Upload GCP L4 GPU-trained sntts Pan-Indic & Sanskrit VITS + 54.3MB Browser ONNX

Browse files
README.md ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - sa
4
+ - hi
5
+ - kn
6
+ - ta
7
+ - te
8
+ - ml
9
+ license: mit
10
+ library_name: onnx
11
+ pipeline_tag: text-to-speech
12
+ tags:
13
+ - text-to-speech
14
+ - sanskrit
15
+ - indic
16
+ - vits
17
+ - hifi-gan
18
+ - onnxruntime-web
19
+ - wasm
20
+ ---
21
+
22
+ # `sntts` — Pan-Indic & Classical Sanskrit Neural TTS (VITS + 16 kHz HiFi-GAN)
23
+
24
+ `sntts` is a compact, single-checkpoint **Pan-Indic & Classical Sanskrit** neural text-to-speech model (`35.4M` parameters; **`54.3 MB` single-file browser ONNX**) supporting **6 Indic languages** (`Sanskrit [sa]`, `Hindi [hi]`, `Kannada [kn]`, `Tamil [ta]`, `Telugu [te]`, `Malayalam [ml]`) across **9 Brahmic scripts + ISO-15919/IAST** in Python, Node.js, and 100% client-side in the browser via `onnxruntime-web` WASM SIMD.
25
+
26
+ ## Training & Evaluation Highlights
27
+ - **GCP GPU Training**: Trained on `NVIDIA L4` (`24` epochs, `3024` steps) across `1,008` studio 16 kHz utterances curated from 7 Hugging Face speech datasets (`prathoshap/sushrota-sanskrit-asr-data`, `yashnbx/gita_supersite_sanskrit_tts`, SYSPIN Hindi/Kannada/Tamil/Telugu, and IITM Malayalam).
28
+ - **Held-Out Evaluation (`N = 126` utterances, `B = 1,000` bootstrap)**:
29
+ - Mean Mel-Cepstral Distortion (`MCD`): **5.143 dB**
30
+ - Independent `Su-śrotā` Sanskrit ASR Round-Trip `CER`: **15.006%**
31
+ - **Files**:
32
+ - `model.safetensors`, `config.json`, `sntts_config.json`, `vocab.json` — PyTorch FP32 checkpoint
33
+ - `onnx/model_quantized.onnx` — Single-file FP16 browser ONNX (`54.3 MB`, zero `.data` sidecars, `r = 0.9998` parity with FP32)
34
+ - `training_history.json`, `gcp_job_manifest.json` — Full training telemetry
config.json ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_attn_implementation_autoset": true,
3
+ "activation_dropout": 0.1,
4
+ "architectures": [
5
+ "VitsModel"
6
+ ],
7
+ "attention_dropout": 0.1,
8
+ "depth_separable_channels": 2,
9
+ "depth_separable_num_layers": 3,
10
+ "dtype": "float32",
11
+ "duration_predictor_dropout": 0.5,
12
+ "duration_predictor_filter_channels": 256,
13
+ "duration_predictor_flow_bins": 10,
14
+ "duration_predictor_kernel_size": 3,
15
+ "duration_predictor_num_flows": 4,
16
+ "duration_predictor_tail_bound": 5.0,
17
+ "ffn_dim": 768,
18
+ "ffn_kernel_size": 3,
19
+ "flow_size": 192,
20
+ "hidden_act": "relu",
21
+ "hidden_dropout": 0.1,
22
+ "hidden_size": 192,
23
+ "initializer_range": 0.02,
24
+ "layer_norm_eps": 1e-05,
25
+ "layerdrop": 0.0,
26
+ "leaky_relu_slope": 0.1,
27
+ "model_type": "vits",
28
+ "noise_scale": 0.667,
29
+ "noise_scale_duration": 0.8,
30
+ "num_attention_heads": 2,
31
+ "num_hidden_layers": 6,
32
+ "num_speakers": 1,
33
+ "pad_token_id": null,
34
+ "posterior_encoder_num_wavenet_layers": 16,
35
+ "prior_encoder_num_flows": 4,
36
+ "prior_encoder_num_wavenet_layers": 4,
37
+ "resblock_dilation_sizes": [
38
+ [
39
+ 1,
40
+ 3,
41
+ 5
42
+ ],
43
+ [
44
+ 1,
45
+ 3,
46
+ 5
47
+ ],
48
+ [
49
+ 1,
50
+ 3,
51
+ 5
52
+ ]
53
+ ],
54
+ "resblock_kernel_sizes": [
55
+ 3,
56
+ 7,
57
+ 11
58
+ ],
59
+ "sampling_rate": 16000,
60
+ "speaker_embedding_size": 0,
61
+ "speaking_rate": 1.0,
62
+ "spectrogram_bins": 513,
63
+ "transformers_version": "5.22.1",
64
+ "upsample_initial_channel": 512,
65
+ "upsample_kernel_sizes": [
66
+ 16,
67
+ 16,
68
+ 4,
69
+ 4
70
+ ],
71
+ "upsample_rates": [
72
+ 8,
73
+ 8,
74
+ 2,
75
+ 2
76
+ ],
77
+ "use_bias": true,
78
+ "use_stochastic_duration_prediction": false,
79
+ "vocab_size": 85,
80
+ "wavenet_dilation_rate": 1,
81
+ "wavenet_dropout": 0.0,
82
+ "wavenet_kernel_size": 5,
83
+ "window_size": 4
84
+ }
gcp_job_manifest.json ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "project": "hemanth-hm",
3
+ "vertex_job_id": "sntts-l4-trainer-wj6r2",
4
+ "region": "us-east4",
5
+ "hostname": "localhost",
6
+ "platform": "Linux-6.9.12-x86_64-with-glibc2.35",
7
+ "pytorch_version": "2.3.1+cu121",
8
+ "cuda_available": true,
9
+ "cuda_version": "12.1",
10
+ "gpu_name": "NVIDIA L4",
11
+ "gpu_vram_gb": 22.03,
12
+ "train_utterances": 1008,
13
+ "val_utterances": 126,
14
+ "epochs": 24,
15
+ "batch_size": 8,
16
+ "total_steps": 3024,
17
+ "best_epoch": 21,
18
+ "total_elapsed_sec": 515.84,
19
+ "parameter_counts": {
20
+ "total_params": 35398659,
21
+ "trainable_params": 13845315,
22
+ "frozen_vocoder_posterior_params": 21553344,
23
+ "head_group_params": 462147,
24
+ "backbone_group_params": 13383168
25
+ },
26
+ "pretrain_baseline": {
27
+ "val_loss": 50.01277,
28
+ "val_mel_l1": 1.28197,
29
+ "val_mr_stft": 2.5412,
30
+ "val_kl_loss": 11.97907,
31
+ "val_dur_loss": 2.06276,
32
+ "val_dur_mae_frames": 4.33768,
33
+ "per_lang": {
34
+ "sa": {
35
+ "n": 36,
36
+ "val_kl": 12.4042,
37
+ "val_dur_mae_frames": 4.4743,
38
+ "val_mel_l1": 1.3366
39
+ },
40
+ "hi": {
41
+ "n": 18,
42
+ "val_kl": 12.4198,
43
+ "val_dur_mae_frames": 4.5223,
44
+ "val_mel_l1": 1.358
45
+ },
46
+ "kn": {
47
+ "n": 18,
48
+ "val_kl": 11.9421,
49
+ "val_dur_mae_frames": 4.3039,
50
+ "val_mel_l1": 1.2303
51
+ },
52
+ "ta": {
53
+ "n": 18,
54
+ "val_kl": 11.6247,
55
+ "val_dur_mae_frames": 4.1909,
56
+ "val_mel_l1": 1.2527
57
+ },
58
+ "te": {
59
+ "n": 18,
60
+ "val_kl": 11.5639,
61
+ "val_dur_mae_frames": 4.2975,
62
+ "val_mel_l1": 1.2351
63
+ },
64
+ "ml": {
65
+ "n": 18,
66
+ "val_kl": 11.4945,
67
+ "val_dur_mae_frames": 4.1006,
68
+ "val_mel_l1": 1.2245
69
+ }
70
+ }
71
+ },
72
+ "best_epoch_metrics": {
73
+ "epoch": 21,
74
+ "step": 2646,
75
+ "epoch_sec": 19.82,
76
+ "lr_head": 5.1e-05,
77
+ "lr_backbone": 1.12e-05,
78
+ "is_best": true,
79
+ "train_loss": 26.59179,
80
+ "train_mel_l1": 0.92518,
81
+ "train_mr_stft": 2.00416,
82
+ "train_kl_loss": 3.88807,
83
+ "train_dur_loss": 0.44005,
84
+ "train_dur_mae_frames": 1.50244,
85
+ "val_loss": 26.87383,
86
+ "val_mel_l1": 0.93228,
87
+ "val_mr_stft": 2.0204,
88
+ "val_kl_loss": 3.98295,
89
+ "val_dur_loss": 0.43507,
90
+ "val_dur_mae_frames": 1.48678,
91
+ "per_lang": {
92
+ "sa": {
93
+ "n": 36,
94
+ "val_kl": 4.0941,
95
+ "val_dur_mae_frames": 1.6749,
96
+ "val_mel_l1": 0.9447
97
+ },
98
+ "hi": {
99
+ "n": 18,
100
+ "val_kl": 4.0062,
101
+ "val_dur_mae_frames": 1.5032,
102
+ "val_mel_l1": 0.9485
103
+ },
104
+ "kn": {
105
+ "n": 18,
106
+ "val_kl": 3.9196,
107
+ "val_dur_mae_frames": 1.5018,
108
+ "val_mel_l1": 0.9178
109
+ },
110
+ "ta": {
111
+ "n": 18,
112
+ "val_kl": 3.8499,
113
+ "val_dur_mae_frames": 1.4425,
114
+ "val_mel_l1": 0.9217
115
+ },
116
+ "te": {
117
+ "n": 18,
118
+ "val_kl": 3.926,
119
+ "val_dur_mae_frames": 1.3312,
120
+ "val_mel_l1": 0.9185
121
+ },
122
+ "ml": {
123
+ "n": 18,
124
+ "val_kl": 3.9907,
125
+ "val_dur_mae_frames": 1.279,
126
+ "val_mel_l1": 0.9302
127
+ }
128
+ }
129
+ },
130
+ "wise_ft": {
131
+ "alpha": 0.35,
132
+ "selection_split": "val_meta.jsonl (N=126, zero test leakage)",
133
+ "val_sweep": {
134
+ "mms_hin_g2p": {
135
+ "val_mcd_raw": 49.56,
136
+ "val_sa_cer_pct": 28.16
137
+ },
138
+ "alpha_0.25": {
139
+ "val_mcd_raw": 43.58,
140
+ "val_sa_cer_pct": 29.24
141
+ },
142
+ "alpha_0.35": {
143
+ "val_mcd_raw": 42.05,
144
+ "val_sa_cer_pct": 27.49
145
+ },
146
+ "alpha_0.50": {
147
+ "val_mcd_raw": 40.4,
148
+ "val_sa_cer_pct": 29.31
149
+ },
150
+ "alpha_0.65": {
151
+ "val_mcd_raw": 39.85,
152
+ "val_sa_cer_pct": 31.41
153
+ },
154
+ "alpha_1.00": {
155
+ "val_mcd_raw": 39.64,
156
+ "val_sa_cer_pct": 44.01
157
+ }
158
+ }
159
+ }
160
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d7d65a9020e9fa1ce29d3e1df1f12689f1ff14b07e986aef78534ae0b6424cfd
3
+ size 141642492
onnx/export_report.json ADDED
@@ -0,0 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "int8_params": 0,
3
+ "fp16_params": 28155968,
4
+ "fp32_params": 16771,
5
+ "fp32_size_mb": 107.93,
6
+ "quantized_size_mb": 54.26,
7
+ "compression_ratio": 1.99,
8
+ "op_types": [
9
+ "Add",
10
+ "Cast",
11
+ "Ceil",
12
+ "Clip",
13
+ "Concat",
14
+ "Constant",
15
+ "ConstantOfShape",
16
+ "Conv",
17
+ "ConvTranspose",
18
+ "CumSum",
19
+ "Div",
20
+ "Exp",
21
+ "Gather",
22
+ "LayerNormalization",
23
+ "LeakyRelu",
24
+ "Less",
25
+ "MatMul",
26
+ "Mul",
27
+ "Neg",
28
+ "Pad",
29
+ "RandomNormalLike",
30
+ "Range",
31
+ "Reciprocal",
32
+ "ReduceMax",
33
+ "ReduceSum",
34
+ "Relu",
35
+ "Reshape",
36
+ "Shape",
37
+ "Sigmoid",
38
+ "Slice",
39
+ "Softmax",
40
+ "Split",
41
+ "Squeeze",
42
+ "Sub",
43
+ "Tanh",
44
+ "Transpose",
45
+ "Unsqueeze"
46
+ ],
47
+ "mean_quant_snr_db": 36.93,
48
+ "min_pearson_r": 0.999185,
49
+ "export_elapsed_sec": 7.85,
50
+ "parity_by_language": [
51
+ {
52
+ "lang": "sa",
53
+ "seq_len": 73,
54
+ "samples": 47360,
55
+ "duration_sec": 2.96,
56
+ "fp32_snr_db": 95.87,
57
+ "quant_snr_db": 38.2,
58
+ "pearson_r": 0.999924,
59
+ "max_abs_diff_fp32": 3.018975257873535e-05
60
+ },
61
+ {
62
+ "lang": "hi",
63
+ "seq_len": 107,
64
+ "samples": 69888,
65
+ "duration_sec": 4.368,
66
+ "fp32_snr_db": 95.99,
67
+ "quant_snr_db": 36.07,
68
+ "pearson_r": 0.999876,
69
+ "max_abs_diff_fp32": 3.83276492357254e-05
70
+ },
71
+ {
72
+ "lang": "kn",
73
+ "seq_len": 71,
74
+ "samples": 54784,
75
+ "duration_sec": 3.424,
76
+ "fp32_snr_db": 99.38,
77
+ "quant_snr_db": 39.95,
78
+ "pearson_r": 0.999949,
79
+ "max_abs_diff_fp32": 2.2932887077331543e-05
80
+ },
81
+ {
82
+ "lang": "ta",
83
+ "seq_len": 51,
84
+ "samples": 33024,
85
+ "duration_sec": 2.064,
86
+ "fp32_snr_db": 96.67,
87
+ "quant_snr_db": 43.37,
88
+ "pearson_r": 0.999977,
89
+ "max_abs_diff_fp32": 4.279613494873047e-05
90
+ },
91
+ {
92
+ "lang": "te",
93
+ "seq_len": 53,
94
+ "samples": 35328,
95
+ "duration_sec": 2.208,
96
+ "fp32_snr_db": 90.0,
97
+ "quant_snr_db": 36.13,
98
+ "pearson_r": 0.999878,
99
+ "max_abs_diff_fp32": 0.00011517107486724854
100
+ },
101
+ {
102
+ "lang": "ml",
103
+ "seq_len": 49,
104
+ "samples": 36096,
105
+ "duration_sec": 2.256,
106
+ "fp32_snr_db": 93.79,
107
+ "quant_snr_db": 27.88,
108
+ "pearson_r": 0.999185,
109
+ "max_abs_diff_fp32": 6.639957427978516e-05
110
+ }
111
+ ]
112
+ }
onnx/model_quantized.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3aff8c3530e6bd28b0c764b96779c28a7815adb91a41bf8cc356b9fb450eddea
3
+ size 56899797
onnx/vocab.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "vocab_tokens": [
3
+ "फ",
4
+ "4",
5
+ "1",
6
+ "-",
7
+ "अ",
8
+ "इ",
9
+ "8",
10
+ " ",
11
+ "0",
12
+ "छ",
13
+ "न",
14
+ "ए",
15
+ "ऐ",
16
+ "़",
17
+ "ष",
18
+ "ि",
19
+ "ँ",
20
+ "च",
21
+ "2",
22
+ "ऑ",
23
+ "थ",
24
+ "भ",
25
+ "ी",
26
+ "‍",
27
+ "ॅ",
28
+ "3",
29
+ "ञ",
30
+ "ै",
31
+ "ु",
32
+ "ठ",
33
+ "ं",
34
+ "ॉ",
35
+ "उ",
36
+ "_",
37
+ "ई",
38
+ "ः",
39
+ "ह",
40
+ "ध",
41
+ "ल",
42
+ "र",
43
+ "स",
44
+ "ब",
45
+ "ख",
46
+ "ण",
47
+ "'",
48
+ "`",
49
+ "व",
50
+ "घ",
51
+ "प",
52
+ "ग",
53
+ "ढ",
54
+ "य",
55
+ "े",
56
+ "्",
57
+ "ा",
58
+ "आ",
59
+ "ड",
60
+ "ज",
61
+ "झ",
62
+ "श",
63
+ "औ",
64
+ "ो",
65
+ "द",
66
+ "ृ",
67
+ "ौ",
68
+ "ऊ",
69
+ "ू",
70
+ "ओ",
71
+ "ट",
72
+ "त",
73
+ "क",
74
+ "म",
75
+ "<unk>",
76
+ "ळ",
77
+ "ऴ",
78
+ "ऱ",
79
+ "ऩ",
80
+ "ऎ",
81
+ "ऒ",
82
+ "ॆ",
83
+ "ॊ",
84
+ "ॠ",
85
+ "ॄ",
86
+ "।",
87
+ "॥"
88
+ ],
89
+ "language_ids": {
90
+ "sa": 0,
91
+ "hi": 1,
92
+ "kn": 2,
93
+ "ta": 3,
94
+ "te": 4,
95
+ "ml": 5
96
+ },
97
+ "sampling_rate": 16000,
98
+ "hop_length": 256
99
+ }
sntts_config.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "vocab_size": 85,
3
+ "num_languages": 6,
4
+ "hidden_size": 192,
5
+ "flow_size": 192,
6
+ "spectrogram_bins": 513,
7
+ "sampling_rate": 16000,
8
+ "n_fft": 1024,
9
+ "hop_length": 256,
10
+ "win_length": 1024,
11
+ "segment_frames": 32,
12
+ "base_checkpoint": "facebook/mms-tts-hin",
13
+ "speaking_rate": 1.0,
14
+ "noise_scale": 0.35
15
+ }
training_history.json ADDED
@@ -0,0 +1,1442 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "epoch": 0,
4
+ "step": 0,
5
+ "val_loss": 50.01277,
6
+ "val_mel_l1": 1.28197,
7
+ "val_mr_stft": 2.5412,
8
+ "val_kl_loss": 11.97907,
9
+ "val_dur_loss": 2.06276,
10
+ "val_dur_mae_frames": 4.33768,
11
+ "per_lang": {
12
+ "sa": {
13
+ "n": 36,
14
+ "val_kl": 12.4042,
15
+ "val_dur_mae_frames": 4.4743,
16
+ "val_mel_l1": 1.3366
17
+ },
18
+ "hi": {
19
+ "n": 18,
20
+ "val_kl": 12.4198,
21
+ "val_dur_mae_frames": 4.5223,
22
+ "val_mel_l1": 1.358
23
+ },
24
+ "kn": {
25
+ "n": 18,
26
+ "val_kl": 11.9421,
27
+ "val_dur_mae_frames": 4.3039,
28
+ "val_mel_l1": 1.2303
29
+ },
30
+ "ta": {
31
+ "n": 18,
32
+ "val_kl": 11.6247,
33
+ "val_dur_mae_frames": 4.1909,
34
+ "val_mel_l1": 1.2527
35
+ },
36
+ "te": {
37
+ "n": 18,
38
+ "val_kl": 11.5639,
39
+ "val_dur_mae_frames": 4.2975,
40
+ "val_mel_l1": 1.2351
41
+ },
42
+ "ml": {
43
+ "n": 18,
44
+ "val_kl": 11.4945,
45
+ "val_dur_mae_frames": 4.1006,
46
+ "val_mel_l1": 1.2245
47
+ }
48
+ }
49
+ },
50
+ {
51
+ "epoch": 1,
52
+ "step": 126,
53
+ "epoch_sec": 31.41,
54
+ "lr_head": 0.0001686,
55
+ "lr_backbone": 3.69e-05,
56
+ "is_best": true,
57
+ "train_loss": 40.3726,
58
+ "train_mel_l1": 1.21104,
59
+ "train_mr_stft": 2.41015,
60
+ "train_kl_loss": 8.34086,
61
+ "train_dur_loss": 0.97896,
62
+ "train_dur_mae_frames": 2.3253,
63
+ "val_loss": 34.97859,
64
+ "val_mel_l1": 1.14804,
65
+ "val_mr_stft": 2.32405,
66
+ "val_kl_loss": 6.2416,
67
+ "val_dur_loss": 0.58617,
68
+ "val_dur_mae_frames": 1.79139,
69
+ "per_lang": {
70
+ "sa": {
71
+ "n": 36,
72
+ "val_kl": 6.4212,
73
+ "val_dur_mae_frames": 1.9767,
74
+ "val_mel_l1": 1.1932
75
+ },
76
+ "hi": {
77
+ "n": 18,
78
+ "val_kl": 6.1823,
79
+ "val_dur_mae_frames": 1.782,
80
+ "val_mel_l1": 1.1562
81
+ },
82
+ "kn": {
83
+ "n": 18,
84
+ "val_kl": 6.2496,
85
+ "val_dur_mae_frames": 1.821,
86
+ "val_mel_l1": 1.1211
87
+ },
88
+ "ta": {
89
+ "n": 18,
90
+ "val_kl": 6.2218,
91
+ "val_dur_mae_frames": 1.7698,
92
+ "val_mel_l1": 1.1457
93
+ },
94
+ "te": {
95
+ "n": 18,
96
+ "val_kl": 6.1355,
97
+ "val_dur_mae_frames": 1.6219,
98
+ "val_mel_l1": 1.0851
99
+ },
100
+ "ml": {
101
+ "n": 18,
102
+ "val_kl": 6.0596,
103
+ "val_dur_mae_frames": 1.5917,
104
+ "val_mel_l1": 1.1418
105
+ }
106
+ }
107
+ },
108
+ {
109
+ "epoch": 2,
110
+ "step": 252,
111
+ "epoch_sec": 23.53,
112
+ "lr_head": 0.00032,
113
+ "lr_backbone": 7e-05,
114
+ "is_best": true,
115
+ "train_loss": 33.5428,
116
+ "train_mel_l1": 1.12337,
117
+ "train_mr_stft": 2.23489,
118
+ "train_kl_loss": 5.61495,
119
+ "train_dur_loss": 0.61896,
120
+ "train_dur_mae_frames": 1.79756,
121
+ "val_loss": 31.56096,
122
+ "val_mel_l1": 1.06839,
123
+ "val_mr_stft": 2.1841,
124
+ "val_kl_loss": 5.11216,
125
+ "val_dur_loss": 0.55419,
126
+ "val_dur_mae_frames": 1.72061,
127
+ "per_lang": {
128
+ "sa": {
129
+ "n": 36,
130
+ "val_kl": 5.2358,
131
+ "val_dur_mae_frames": 1.9167,
132
+ "val_mel_l1": 1.0916
133
+ },
134
+ "hi": {
135
+ "n": 18,
136
+ "val_kl": 4.9843,
137
+ "val_dur_mae_frames": 1.6981,
138
+ "val_mel_l1": 1.0657
139
+ },
140
+ "kn": {
141
+ "n": 18,
142
+ "val_kl": 5.187,
143
+ "val_dur_mae_frames": 1.7703,
144
+ "val_mel_l1": 1.0519
145
+ },
146
+ "ta": {
147
+ "n": 18,
148
+ "val_kl": 5.0016,
149
+ "val_dur_mae_frames": 1.6787,
150
+ "val_mel_l1": 1.07
151
+ },
152
+ "te": {
153
+ "n": 18,
154
+ "val_kl": 5.1286,
155
+ "val_dur_mae_frames": 1.5608,
156
+ "val_mel_l1": 1.0433
157
+ },
158
+ "ml": {
159
+ "n": 18,
160
+ "val_kl": 5.012,
161
+ "val_dur_mae_frames": 1.5029,
162
+ "val_mel_l1": 1.0645
163
+ }
164
+ }
165
+ },
166
+ {
167
+ "epoch": 3,
168
+ "step": 378,
169
+ "epoch_sec": 22.69,
170
+ "lr_head": 0.0003183,
171
+ "lr_backbone": 6.96e-05,
172
+ "is_best": false,
173
+ "train_loss": 32.12078,
174
+ "train_mel_l1": 1.08712,
175
+ "train_mr_stft": 2.19205,
176
+ "train_kl_loss": 5.2154,
177
+ "train_dur_loss": 0.57659,
178
+ "train_dur_mae_frames": 1.74254,
179
+ "val_loss": 31.94791,
180
+ "val_mel_l1": 1.0939,
181
+ "val_mr_stft": 2.20019,
182
+ "val_kl_loss": 5.03934,
183
+ "val_dur_loss": 0.55939,
184
+ "val_dur_mae_frames": 1.73812,
185
+ "per_lang": {
186
+ "sa": {
187
+ "n": 36,
188
+ "val_kl": 5.1152,
189
+ "val_dur_mae_frames": 1.9324,
190
+ "val_mel_l1": 1.1223
191
+ },
192
+ "hi": {
193
+ "n": 18,
194
+ "val_kl": 5.0649,
195
+ "val_dur_mae_frames": 1.7402,
196
+ "val_mel_l1": 1.1165
197
+ },
198
+ "kn": {
199
+ "n": 18,
200
+ "val_kl": 5.0144,
201
+ "val_dur_mae_frames": 1.7693,
202
+ "val_mel_l1": 1.1035
203
+ },
204
+ "ta": {
205
+ "n": 18,
206
+ "val_kl": 4.9701,
207
+ "val_dur_mae_frames": 1.6866,
208
+ "val_mel_l1": 1.0413
209
+ },
210
+ "te": {
211
+ "n": 18,
212
+ "val_kl": 4.9521,
213
+ "val_dur_mae_frames": 1.5779,
214
+ "val_mel_l1": 1.0773
215
+ },
216
+ "ml": {
217
+ "n": 18,
218
+ "val_kl": 5.0435,
219
+ "val_dur_mae_frames": 1.5279,
220
+ "val_mel_l1": 1.0739
221
+ }
222
+ }
223
+ },
224
+ {
225
+ "epoch": 4,
226
+ "step": 504,
227
+ "epoch_sec": 21.82,
228
+ "lr_head": 0.0003138,
229
+ "lr_backbone": 6.87e-05,
230
+ "is_best": true,
231
+ "train_loss": 31.04811,
232
+ "train_mel_l1": 1.06097,
233
+ "train_mr_stft": 2.15762,
234
+ "train_kl_loss": 4.89917,
235
+ "train_dur_loss": 0.54622,
236
+ "train_dur_mae_frames": 1.69315,
237
+ "val_loss": 30.83196,
238
+ "val_mel_l1": 1.06086,
239
+ "val_mr_stft": 2.15814,
240
+ "val_kl_loss": 4.8058,
241
+ "val_dur_loss": 0.51622,
242
+ "val_dur_mae_frames": 1.66177,
243
+ "per_lang": {
244
+ "sa": {
245
+ "n": 36,
246
+ "val_kl": 4.8476,
247
+ "val_dur_mae_frames": 1.8473,
248
+ "val_mel_l1": 1.083
249
+ },
250
+ "hi": {
251
+ "n": 18,
252
+ "val_kl": 4.7803,
253
+ "val_dur_mae_frames": 1.6579,
254
+ "val_mel_l1": 1.0498
255
+ },
256
+ "kn": {
257
+ "n": 18,
258
+ "val_kl": 4.679,
259
+ "val_dur_mae_frames": 1.7187,
260
+ "val_mel_l1": 1.0363
261
+ },
262
+ "ta": {
263
+ "n": 18,
264
+ "val_kl": 4.6565,
265
+ "val_dur_mae_frames": 1.6147,
266
+ "val_mel_l1": 1.0483
267
+ },
268
+ "te": {
269
+ "n": 18,
270
+ "val_kl": 4.8405,
271
+ "val_dur_mae_frames": 1.5046,
272
+ "val_mel_l1": 1.0537
273
+ },
274
+ "ml": {
275
+ "n": 18,
276
+ "val_kl": 4.989,
277
+ "val_dur_mae_frames": 1.4419,
278
+ "val_mel_l1": 1.0719
279
+ }
280
+ }
281
+ },
282
+ {
283
+ "epoch": 5,
284
+ "step": 630,
285
+ "epoch_sec": 21.1,
286
+ "lr_head": 0.0003066,
287
+ "lr_backbone": 6.71e-05,
288
+ "is_best": true,
289
+ "train_loss": 30.37543,
290
+ "train_mel_l1": 1.04361,
291
+ "train_mr_stft": 2.14098,
292
+ "train_kl_loss": 4.70154,
293
+ "train_dur_loss": 0.53066,
294
+ "train_dur_mae_frames": 1.66799,
295
+ "val_loss": 30.32108,
296
+ "val_mel_l1": 1.04168,
297
+ "val_mr_stft": 2.12082,
298
+ "val_kl_loss": 4.74305,
299
+ "val_dur_loss": 0.51005,
300
+ "val_dur_mae_frames": 1.60804,
301
+ "per_lang": {
302
+ "sa": {
303
+ "n": 36,
304
+ "val_kl": 4.84,
305
+ "val_dur_mae_frames": 1.7981,
306
+ "val_mel_l1": 1.077
307
+ },
308
+ "hi": {
309
+ "n": 18,
310
+ "val_kl": 4.7413,
311
+ "val_dur_mae_frames": 1.5954,
312
+ "val_mel_l1": 1.0592
313
+ },
314
+ "kn": {
315
+ "n": 18,
316
+ "val_kl": 4.7121,
317
+ "val_dur_mae_frames": 1.6655,
318
+ "val_mel_l1": 0.9947
319
+ },
320
+ "ta": {
321
+ "n": 18,
322
+ "val_kl": 4.5461,
323
+ "val_dur_mae_frames": 1.5696,
324
+ "val_mel_l1": 1.0209
325
+ },
326
+ "te": {
327
+ "n": 18,
328
+ "val_kl": 4.8299,
329
+ "val_dur_mae_frames": 1.4437,
330
+ "val_mel_l1": 1.0392
331
+ },
332
+ "ml": {
333
+ "n": 18,
334
+ "val_kl": 4.6919,
335
+ "val_dur_mae_frames": 1.3859,
336
+ "val_mel_l1": 1.0237
337
+ }
338
+ }
339
+ },
340
+ {
341
+ "epoch": 6,
342
+ "step": 756,
343
+ "epoch_sec": 20.53,
344
+ "lr_head": 0.0002969,
345
+ "lr_backbone": 6.49e-05,
346
+ "is_best": false,
347
+ "train_loss": 30.11437,
348
+ "train_mel_l1": 1.03573,
349
+ "train_mr_stft": 2.13768,
350
+ "train_kl_loss": 4.65741,
351
+ "train_dur_loss": 0.51141,
352
+ "train_dur_mae_frames": 1.63141,
353
+ "val_loss": 30.56487,
354
+ "val_mel_l1": 1.05872,
355
+ "val_mr_stft": 2.13312,
356
+ "val_kl_loss": 4.68022,
357
+ "val_dur_loss": 0.51516,
358
+ "val_dur_mae_frames": 1.64743,
359
+ "per_lang": {
360
+ "sa": {
361
+ "n": 36,
362
+ "val_kl": 4.8208,
363
+ "val_dur_mae_frames": 1.842,
364
+ "val_mel_l1": 1.0771
365
+ },
366
+ "hi": {
367
+ "n": 18,
368
+ "val_kl": 4.6105,
369
+ "val_dur_mae_frames": 1.6486,
370
+ "val_mel_l1": 1.0672
371
+ },
372
+ "kn": {
373
+ "n": 18,
374
+ "val_kl": 4.6241,
375
+ "val_dur_mae_frames": 1.6748,
376
+ "val_mel_l1": 1.0607
377
+ },
378
+ "ta": {
379
+ "n": 18,
380
+ "val_kl": 4.4796,
381
+ "val_dur_mae_frames": 1.603,
382
+ "val_mel_l1": 1.0265
383
+ },
384
+ "te": {
385
+ "n": 18,
386
+ "val_kl": 4.739,
387
+ "val_dur_mae_frames": 1.489,
388
+ "val_mel_l1": 1.0484
389
+ },
390
+ "ml": {
391
+ "n": 18,
392
+ "val_kl": 4.6668,
393
+ "val_dur_mae_frames": 1.4326,
394
+ "val_mel_l1": 1.054
395
+ }
396
+ }
397
+ },
398
+ {
399
+ "epoch": 7,
400
+ "step": 882,
401
+ "epoch_sec": 21.02,
402
+ "lr_head": 0.0002847,
403
+ "lr_backbone": 6.23e-05,
404
+ "is_best": true,
405
+ "train_loss": 29.32643,
406
+ "train_mel_l1": 1.00772,
407
+ "train_mr_stft": 2.10406,
408
+ "train_kl_loss": 4.5126,
409
+ "train_dur_loss": 0.50495,
410
+ "train_dur_mae_frames": 1.62458,
411
+ "val_loss": 29.81836,
412
+ "val_mel_l1": 1.03498,
413
+ "val_mr_stft": 2.12574,
414
+ "val_kl_loss": 4.52396,
415
+ "val_dur_loss": 0.48566,
416
+ "val_dur_mae_frames": 1.58604,
417
+ "per_lang": {
418
+ "sa": {
419
+ "n": 36,
420
+ "val_kl": 4.6419,
421
+ "val_dur_mae_frames": 1.7694,
422
+ "val_mel_l1": 1.0636
423
+ },
424
+ "hi": {
425
+ "n": 18,
426
+ "val_kl": 4.4526,
427
+ "val_dur_mae_frames": 1.5875,
428
+ "val_mel_l1": 1.032
429
+ },
430
+ "kn": {
431
+ "n": 18,
432
+ "val_kl": 4.3727,
433
+ "val_dur_mae_frames": 1.6185,
434
+ "val_mel_l1": 1.007
435
+ },
436
+ "ta": {
437
+ "n": 18,
438
+ "val_kl": 4.4886,
439
+ "val_dur_mae_frames": 1.5485,
440
+ "val_mel_l1": 1.0135
441
+ },
442
+ "te": {
443
+ "n": 18,
444
+ "val_kl": 4.4551,
445
+ "val_dur_mae_frames": 1.4306,
446
+ "val_mel_l1": 1.0162
447
+ },
448
+ "ml": {
449
+ "n": 18,
450
+ "val_kl": 4.615,
451
+ "val_dur_mae_frames": 1.3783,
452
+ "val_mel_l1": 1.049
453
+ }
454
+ }
455
+ },
456
+ {
457
+ "epoch": 8,
458
+ "step": 1008,
459
+ "epoch_sec": 20.35,
460
+ "lr_head": 0.0002704,
461
+ "lr_backbone": 5.92e-05,
462
+ "is_best": true,
463
+ "train_loss": 28.97411,
464
+ "train_mel_l1": 1.00004,
465
+ "train_mr_stft": 2.09019,
466
+ "train_kl_loss": 4.39955,
467
+ "train_dur_loss": 0.49548,
468
+ "train_dur_mae_frames": 1.60589,
469
+ "val_loss": 29.05696,
470
+ "val_mel_l1": 0.99922,
471
+ "val_mr_stft": 2.08642,
472
+ "val_kl_loss": 4.50708,
473
+ "val_dur_loss": 0.47228,
474
+ "val_dur_mae_frames": 1.55318,
475
+ "per_lang": {
476
+ "sa": {
477
+ "n": 36,
478
+ "val_kl": 4.6074,
479
+ "val_dur_mae_frames": 1.7377,
480
+ "val_mel_l1": 1.0067
481
+ },
482
+ "hi": {
483
+ "n": 18,
484
+ "val_kl": 4.4562,
485
+ "val_dur_mae_frames": 1.5481,
486
+ "val_mel_l1": 1.0103
487
+ },
488
+ "kn": {
489
+ "n": 18,
490
+ "val_kl": 4.4426,
491
+ "val_dur_mae_frames": 1.5937,
492
+ "val_mel_l1": 0.9908
493
+ },
494
+ "ta": {
495
+ "n": 18,
496
+ "val_kl": 4.3434,
497
+ "val_dur_mae_frames": 1.5185,
498
+ "val_mel_l1": 0.9736
499
+ },
500
+ "te": {
501
+ "n": 18,
502
+ "val_kl": 4.5093,
503
+ "val_dur_mae_frames": 1.3922,
504
+ "val_mel_l1": 0.9848
505
+ },
506
+ "ml": {
507
+ "n": 18,
508
+ "val_kl": 4.5832,
509
+ "val_dur_mae_frames": 1.3443,
510
+ "val_mel_l1": 1.0216
511
+ }
512
+ }
513
+ },
514
+ {
515
+ "epoch": 9,
516
+ "step": 1134,
517
+ "epoch_sec": 19.91,
518
+ "lr_head": 0.0002543,
519
+ "lr_backbone": 5.56e-05,
520
+ "is_best": true,
521
+ "train_loss": 28.42117,
522
+ "train_mel_l1": 0.98124,
523
+ "train_mr_stft": 2.06698,
524
+ "train_kl_loss": 4.30254,
525
+ "train_dur_loss": 0.48182,
526
+ "train_dur_mae_frames": 1.57749,
527
+ "val_loss": 28.88793,
528
+ "val_mel_l1": 0.9987,
529
+ "val_mr_stft": 2.08552,
530
+ "val_kl_loss": 4.41872,
531
+ "val_dur_loss": 0.46195,
532
+ "val_dur_mae_frames": 1.55001,
533
+ "per_lang": {
534
+ "sa": {
535
+ "n": 36,
536
+ "val_kl": 4.4646,
537
+ "val_dur_mae_frames": 1.7314,
538
+ "val_mel_l1": 1.0239
539
+ },
540
+ "hi": {
541
+ "n": 18,
542
+ "val_kl": 4.291,
543
+ "val_dur_mae_frames": 1.5501,
544
+ "val_mel_l1": 0.9941
545
+ },
546
+ "kn": {
547
+ "n": 18,
548
+ "val_kl": 4.3949,
549
+ "val_dur_mae_frames": 1.5667,
550
+ "val_mel_l1": 0.969
551
+ },
552
+ "ta": {
553
+ "n": 18,
554
+ "val_kl": 4.3421,
555
+ "val_dur_mae_frames": 1.5202,
556
+ "val_mel_l1": 0.9819
557
+ },
558
+ "te": {
559
+ "n": 18,
560
+ "val_kl": 4.4758,
561
+ "val_dur_mae_frames": 1.3964,
562
+ "val_mel_l1": 0.9786
563
+ },
564
+ "ml": {
565
+ "n": 18,
566
+ "val_kl": 4.498,
567
+ "val_dur_mae_frames": 1.3539,
568
+ "val_mel_l1": 1.0194
569
+ }
570
+ }
571
+ },
572
+ {
573
+ "epoch": 10,
574
+ "step": 1260,
575
+ "epoch_sec": 20.06,
576
+ "lr_head": 0.0002367,
577
+ "lr_backbone": 5.18e-05,
578
+ "is_best": true,
579
+ "train_loss": 28.38448,
580
+ "train_mel_l1": 0.97604,
581
+ "train_mr_stft": 2.06446,
582
+ "train_kl_loss": 4.34999,
583
+ "train_dur_loss": 0.47764,
584
+ "train_dur_mae_frames": 1.5709,
585
+ "val_loss": 28.74381,
586
+ "val_mel_l1": 0.98578,
587
+ "val_mr_stft": 2.0821,
588
+ "val_kl_loss": 4.47202,
589
+ "val_dur_loss": 0.46747,
590
+ "val_dur_mae_frames": 1.55994,
591
+ "per_lang": {
592
+ "sa": {
593
+ "n": 36,
594
+ "val_kl": 4.5571,
595
+ "val_dur_mae_frames": 1.7443,
596
+ "val_mel_l1": 1.0116
597
+ },
598
+ "hi": {
599
+ "n": 18,
600
+ "val_kl": 4.3889,
601
+ "val_dur_mae_frames": 1.5647,
602
+ "val_mel_l1": 0.9801
603
+ },
604
+ "kn": {
605
+ "n": 18,
606
+ "val_kl": 4.4193,
607
+ "val_dur_mae_frames": 1.5854,
608
+ "val_mel_l1": 1.0002
609
+ },
610
+ "ta": {
611
+ "n": 18,
612
+ "val_kl": 4.3948,
613
+ "val_dur_mae_frames": 1.5254,
614
+ "val_mel_l1": 0.9494
615
+ },
616
+ "te": {
617
+ "n": 18,
618
+ "val_kl": 4.4334,
619
+ "val_dur_mae_frames": 1.4061,
620
+ "val_mel_l1": 0.9776
621
+ },
622
+ "ml": {
623
+ "n": 18,
624
+ "val_kl": 4.5536,
625
+ "val_dur_mae_frames": 1.3495,
626
+ "val_mel_l1": 0.9698
627
+ }
628
+ }
629
+ },
630
+ {
631
+ "epoch": 11,
632
+ "step": 1386,
633
+ "epoch_sec": 19.92,
634
+ "lr_head": 0.0002179,
635
+ "lr_backbone": 4.77e-05,
636
+ "is_best": true,
637
+ "train_loss": 28.13614,
638
+ "train_mel_l1": 0.97335,
639
+ "train_mr_stft": 2.06204,
640
+ "train_kl_loss": 4.23002,
641
+ "train_dur_loss": 0.47107,
642
+ "train_dur_mae_frames": 1.55871,
643
+ "val_loss": 28.28355,
644
+ "val_mel_l1": 0.98302,
645
+ "val_mr_stft": 2.06611,
646
+ "val_kl_loss": 4.22469,
647
+ "val_dur_loss": 0.4612,
648
+ "val_dur_mae_frames": 1.54911,
649
+ "per_lang": {
650
+ "sa": {
651
+ "n": 36,
652
+ "val_kl": 4.3386,
653
+ "val_dur_mae_frames": 1.7288,
654
+ "val_mel_l1": 1.0057
655
+ },
656
+ "hi": {
657
+ "n": 18,
658
+ "val_kl": 4.1166,
659
+ "val_dur_mae_frames": 1.5513,
660
+ "val_mel_l1": 0.983
661
+ },
662
+ "kn": {
663
+ "n": 18,
664
+ "val_kl": 4.1309,
665
+ "val_dur_mae_frames": 1.5823,
666
+ "val_mel_l1": 0.9851
667
+ },
668
+ "ta": {
669
+ "n": 18,
670
+ "val_kl": 4.1116,
671
+ "val_dur_mae_frames": 1.5177,
672
+ "val_mel_l1": 0.958
673
+ },
674
+ "te": {
675
+ "n": 18,
676
+ "val_kl": 4.2318,
677
+ "val_dur_mae_frames": 1.3936,
678
+ "val_mel_l1": 0.9579
679
+ },
680
+ "ml": {
681
+ "n": 18,
682
+ "val_kl": 4.3047,
683
+ "val_dur_mae_frames": 1.3413,
684
+ "val_mel_l1": 0.9858
685
+ }
686
+ }
687
+ },
688
+ {
689
+ "epoch": 12,
690
+ "step": 1512,
691
+ "epoch_sec": 19.81,
692
+ "lr_head": 0.0001983,
693
+ "lr_backbone": 4.34e-05,
694
+ "is_best": true,
695
+ "train_loss": 27.73874,
696
+ "train_mel_l1": 0.95815,
697
+ "train_mr_stft": 2.03953,
698
+ "train_kl_loss": 4.17482,
699
+ "train_dur_loss": 0.4682,
700
+ "train_dur_mae_frames": 1.5548,
701
+ "val_loss": 27.89086,
702
+ "val_mel_l1": 0.96948,
703
+ "val_mr_stft": 2.04298,
704
+ "val_kl_loss": 4.16125,
705
+ "val_dur_loss": 0.45356,
706
+ "val_dur_mae_frames": 1.51624,
707
+ "per_lang": {
708
+ "sa": {
709
+ "n": 36,
710
+ "val_kl": 4.2201,
711
+ "val_dur_mae_frames": 1.7003,
712
+ "val_mel_l1": 0.996
713
+ },
714
+ "hi": {
715
+ "n": 18,
716
+ "val_kl": 4.1633,
717
+ "val_dur_mae_frames": 1.5134,
718
+ "val_mel_l1": 0.9795
719
+ },
720
+ "kn": {
721
+ "n": 18,
722
+ "val_kl": 4.0876,
723
+ "val_dur_mae_frames": 1.5456,
724
+ "val_mel_l1": 0.9473
725
+ },
726
+ "ta": {
727
+ "n": 18,
728
+ "val_kl": 4.0319,
729
+ "val_dur_mae_frames": 1.478,
730
+ "val_mel_l1": 0.9454
731
+ },
732
+ "te": {
733
+ "n": 18,
734
+ "val_kl": 4.1853,
735
+ "val_dur_mae_frames": 1.3637,
736
+ "val_mel_l1": 0.956
737
+ },
738
+ "ml": {
739
+ "n": 18,
740
+ "val_kl": 4.2204,
741
+ "val_dur_mae_frames": 1.3126,
742
+ "val_mel_l1": 0.9661
743
+ }
744
+ }
745
+ },
746
+ {
747
+ "epoch": 13,
748
+ "step": 1638,
749
+ "epoch_sec": 19.9,
750
+ "lr_head": 0.0001783,
751
+ "lr_backbone": 3.9e-05,
752
+ "is_best": false,
753
+ "train_loss": 27.63173,
754
+ "train_mel_l1": 0.95826,
755
+ "train_mr_stft": 2.04231,
756
+ "train_kl_loss": 4.10793,
757
+ "train_dur_loss": 0.46305,
758
+ "train_dur_mae_frames": 1.54321,
759
+ "val_loss": 27.87036,
760
+ "val_mel_l1": 0.96385,
761
+ "val_mr_stft": 2.03278,
762
+ "val_kl_loss": 4.23105,
763
+ "val_dur_loss": 0.45011,
764
+ "val_dur_mae_frames": 1.52217,
765
+ "per_lang": {
766
+ "sa": {
767
+ "n": 36,
768
+ "val_kl": 4.3136,
769
+ "val_dur_mae_frames": 1.7095,
770
+ "val_mel_l1": 0.9813
771
+ },
772
+ "hi": {
773
+ "n": 18,
774
+ "val_kl": 4.1001,
775
+ "val_dur_mae_frames": 1.5318,
776
+ "val_mel_l1": 0.953
777
+ },
778
+ "kn": {
779
+ "n": 18,
780
+ "val_kl": 4.2369,
781
+ "val_dur_mae_frames": 1.5355,
782
+ "val_mel_l1": 0.9773
783
+ },
784
+ "ta": {
785
+ "n": 18,
786
+ "val_kl": 4.092,
787
+ "val_dur_mae_frames": 1.4918,
788
+ "val_mel_l1": 0.9443
789
+ },
790
+ "te": {
791
+ "n": 18,
792
+ "val_kl": 4.262,
793
+ "val_dur_mae_frames": 1.359,
794
+ "val_mel_l1": 0.9571
795
+ },
796
+ "ml": {
797
+ "n": 18,
798
+ "val_kl": 4.2993,
799
+ "val_dur_mae_frames": 1.318,
800
+ "val_mel_l1": 0.9526
801
+ }
802
+ }
803
+ },
804
+ {
805
+ "epoch": 14,
806
+ "step": 1764,
807
+ "epoch_sec": 19.53,
808
+ "lr_head": 0.0001584,
809
+ "lr_backbone": 3.46e-05,
810
+ "is_best": false,
811
+ "train_loss": 27.35702,
812
+ "train_mel_l1": 0.94663,
813
+ "train_mr_stft": 2.0261,
814
+ "train_kl_loss": 4.08487,
815
+ "train_dur_loss": 0.46048,
816
+ "train_dur_mae_frames": 1.54047,
817
+ "val_loss": 28.17127,
818
+ "val_mel_l1": 0.98632,
819
+ "val_mr_stft": 2.05798,
820
+ "val_kl_loss": 4.13878,
821
+ "val_dur_loss": 0.44892,
822
+ "val_dur_mae_frames": 1.50657,
823
+ "per_lang": {
824
+ "sa": {
825
+ "n": 36,
826
+ "val_kl": 4.2212,
827
+ "val_dur_mae_frames": 1.6902,
828
+ "val_mel_l1": 1.0073
829
+ },
830
+ "hi": {
831
+ "n": 18,
832
+ "val_kl": 4.0138,
833
+ "val_dur_mae_frames": 1.5145,
834
+ "val_mel_l1": 0.9762
835
+ },
836
+ "kn": {
837
+ "n": 18,
838
+ "val_kl": 4.0224,
839
+ "val_dur_mae_frames": 1.5205,
840
+ "val_mel_l1": 0.9772
841
+ },
842
+ "ta": {
843
+ "n": 18,
844
+ "val_kl": 4.0084,
845
+ "val_dur_mae_frames": 1.4742,
846
+ "val_mel_l1": 0.9497
847
+ },
848
+ "te": {
849
+ "n": 18,
850
+ "val_kl": 4.1897,
851
+ "val_dur_mae_frames": 1.3471,
852
+ "val_mel_l1": 0.9824
853
+ },
854
+ "ml": {
855
+ "n": 18,
856
+ "val_kl": 4.2946,
857
+ "val_dur_mae_frames": 1.3092,
858
+ "val_mel_l1": 1.0041
859
+ }
860
+ }
861
+ },
862
+ {
863
+ "epoch": 15,
864
+ "step": 1890,
865
+ "epoch_sec": 19.71,
866
+ "lr_head": 0.0001388,
867
+ "lr_backbone": 3.04e-05,
868
+ "is_best": true,
869
+ "train_loss": 27.19794,
870
+ "train_mel_l1": 0.94089,
871
+ "train_mr_stft": 2.02056,
872
+ "train_kl_loss": 4.05353,
873
+ "train_dur_loss": 0.46035,
874
+ "train_dur_mae_frames": 1.54007,
875
+ "val_loss": 27.76906,
876
+ "val_mel_l1": 0.96549,
877
+ "val_mr_stft": 2.04674,
878
+ "val_kl_loss": 4.12817,
879
+ "val_dur_loss": 0.45113,
880
+ "val_dur_mae_frames": 1.5258,
881
+ "per_lang": {
882
+ "sa": {
883
+ "n": 36,
884
+ "val_kl": 4.1878,
885
+ "val_dur_mae_frames": 1.6998,
886
+ "val_mel_l1": 0.974
887
+ },
888
+ "hi": {
889
+ "n": 18,
890
+ "val_kl": 4.134,
891
+ "val_dur_mae_frames": 1.5403,
892
+ "val_mel_l1": 0.9807
893
+ },
894
+ "kn": {
895
+ "n": 18,
896
+ "val_kl": 3.9863,
897
+ "val_dur_mae_frames": 1.5465,
898
+ "val_mel_l1": 0.9709
899
+ },
900
+ "ta": {
901
+ "n": 18,
902
+ "val_kl": 4.042,
903
+ "val_dur_mae_frames": 1.4853,
904
+ "val_mel_l1": 0.9234
905
+ },
906
+ "te": {
907
+ "n": 18,
908
+ "val_kl": 4.2153,
909
+ "val_dur_mae_frames": 1.375,
910
+ "val_mel_l1": 0.9723
911
+ },
912
+ "ml": {
913
+ "n": 18,
914
+ "val_kl": 4.144,
915
+ "val_dur_mae_frames": 1.3339,
916
+ "val_mel_l1": 0.9631
917
+ }
918
+ }
919
+ },
920
+ {
921
+ "epoch": 16,
922
+ "step": 2016,
923
+ "epoch_sec": 19.44,
924
+ "lr_head": 0.0001201,
925
+ "lr_backbone": 2.63e-05,
926
+ "is_best": false,
927
+ "train_loss": 27.07208,
928
+ "train_mel_l1": 0.93817,
929
+ "train_mr_stft": 2.01902,
930
+ "train_kl_loss": 4.01455,
931
+ "train_dur_loss": 0.45384,
932
+ "train_dur_mae_frames": 1.52551,
933
+ "val_loss": 28.05535,
934
+ "val_mel_l1": 0.98052,
935
+ "val_mr_stft": 2.06771,
936
+ "val_kl_loss": 4.12877,
937
+ "val_dur_loss": 0.44451,
938
+ "val_dur_mae_frames": 1.49581,
939
+ "per_lang": {
940
+ "sa": {
941
+ "n": 36,
942
+ "val_kl": 4.2247,
943
+ "val_dur_mae_frames": 1.6715,
944
+ "val_mel_l1": 1.0039
945
+ },
946
+ "hi": {
947
+ "n": 18,
948
+ "val_kl": 3.9948,
949
+ "val_dur_mae_frames": 1.5023,
950
+ "val_mel_l1": 0.9905
951
+ },
952
+ "kn": {
953
+ "n": 18,
954
+ "val_kl": 3.9838,
955
+ "val_dur_mae_frames": 1.5136,
956
+ "val_mel_l1": 0.9679
957
+ },
958
+ "ta": {
959
+ "n": 18,
960
+ "val_kl": 4.14,
961
+ "val_dur_mae_frames": 1.4639,
962
+ "val_mel_l1": 0.9549
963
+ },
964
+ "te": {
965
+ "n": 18,
966
+ "val_kl": 4.1015,
967
+ "val_dur_mae_frames": 1.3425,
968
+ "val_mel_l1": 0.9567
969
+ },
970
+ "ml": {
971
+ "n": 18,
972
+ "val_kl": 4.2319,
973
+ "val_dur_mae_frames": 1.3053,
974
+ "val_mel_l1": 0.9857
975
+ }
976
+ }
977
+ },
978
+ {
979
+ "epoch": 17,
980
+ "step": 2142,
981
+ "epoch_sec": 19.67,
982
+ "lr_head": 0.0001026,
983
+ "lr_backbone": 2.24e-05,
984
+ "is_best": true,
985
+ "train_loss": 27.09273,
986
+ "train_mel_l1": 0.93927,
987
+ "train_mr_stft": 2.02333,
988
+ "train_kl_loss": 4.0233,
989
+ "train_dur_loss": 0.4464,
990
+ "train_dur_mae_frames": 1.51136,
991
+ "val_loss": 27.25504,
992
+ "val_mel_l1": 0.95394,
993
+ "val_mr_stft": 2.02004,
994
+ "val_kl_loss": 3.97197,
995
+ "val_dur_loss": 0.43848,
996
+ "val_dur_mae_frames": 1.49053,
997
+ "per_lang": {
998
+ "sa": {
999
+ "n": 36,
1000
+ "val_kl": 4.0633,
1001
+ "val_dur_mae_frames": 1.6749,
1002
+ "val_mel_l1": 0.9794
1003
+ },
1004
+ "hi": {
1005
+ "n": 18,
1006
+ "val_kl": 3.9402,
1007
+ "val_dur_mae_frames": 1.5063,
1008
+ "val_mel_l1": 0.928
1009
+ },
1010
+ "kn": {
1011
+ "n": 18,
1012
+ "val_kl": 3.9188,
1013
+ "val_dur_mae_frames": 1.5021,
1014
+ "val_mel_l1": 0.9613
1015
+ },
1016
+ "ta": {
1017
+ "n": 18,
1018
+ "val_kl": 3.8537,
1019
+ "val_dur_mae_frames": 1.457,
1020
+ "val_mel_l1": 0.9345
1021
+ },
1022
+ "te": {
1023
+ "n": 18,
1024
+ "val_kl": 3.934,
1025
+ "val_dur_mae_frames": 1.3383,
1026
+ "val_mel_l1": 0.9228
1027
+ },
1028
+ "ml": {
1029
+ "n": 18,
1030
+ "val_kl": 4.0304,
1031
+ "val_dur_mae_frames": 1.2803,
1032
+ "val_mel_l1": 0.9722
1033
+ }
1034
+ }
1035
+ },
1036
+ {
1037
+ "epoch": 18,
1038
+ "step": 2268,
1039
+ "epoch_sec": 19.72,
1040
+ "lr_head": 8.66e-05,
1041
+ "lr_backbone": 1.9e-05,
1042
+ "is_best": false,
1043
+ "train_loss": 26.87977,
1044
+ "train_mel_l1": 0.9326,
1045
+ "train_mr_stft": 2.01281,
1046
+ "train_kl_loss": 3.97522,
1047
+ "train_dur_loss": 0.44437,
1048
+ "train_dur_mae_frames": 1.5069,
1049
+ "val_loss": 27.6553,
1050
+ "val_mel_l1": 0.96327,
1051
+ "val_mr_stft": 2.03311,
1052
+ "val_kl_loss": 4.11443,
1053
+ "val_dur_loss": 0.43807,
1054
+ "val_dur_mae_frames": 1.49109,
1055
+ "per_lang": {
1056
+ "sa": {
1057
+ "n": 36,
1058
+ "val_kl": 4.1253,
1059
+ "val_dur_mae_frames": 1.675,
1060
+ "val_mel_l1": 0.998
1061
+ },
1062
+ "hi": {
1063
+ "n": 18,
1064
+ "val_kl": 4.1125,
1065
+ "val_dur_mae_frames": 1.5056,
1066
+ "val_mel_l1": 0.9348
1067
+ },
1068
+ "kn": {
1069
+ "n": 18,
1070
+ "val_kl": 4.0268,
1071
+ "val_dur_mae_frames": 1.5055,
1072
+ "val_mel_l1": 0.9828
1073
+ },
1074
+ "ta": {
1075
+ "n": 18,
1076
+ "val_kl": 4.104,
1077
+ "val_dur_mae_frames": 1.455,
1078
+ "val_mel_l1": 0.9545
1079
+ },
1080
+ "te": {
1081
+ "n": 18,
1082
+ "val_kl": 4.0839,
1083
+ "val_dur_mae_frames": 1.3374,
1084
+ "val_mel_l1": 0.9321
1085
+ },
1086
+ "ml": {
1087
+ "n": 18,
1088
+ "val_kl": 4.2232,
1089
+ "val_dur_mae_frames": 1.2841,
1090
+ "val_mel_l1": 0.9426
1091
+ }
1092
+ }
1093
+ },
1094
+ {
1095
+ "epoch": 19,
1096
+ "step": 2394,
1097
+ "epoch_sec": 19.84,
1098
+ "lr_head": 7.25e-05,
1099
+ "lr_backbone": 1.59e-05,
1100
+ "is_best": false,
1101
+ "train_loss": 26.80541,
1102
+ "train_mel_l1": 0.93106,
1103
+ "train_mr_stft": 2.01271,
1104
+ "train_kl_loss": 3.94564,
1105
+ "train_dur_loss": 0.44355,
1106
+ "train_dur_mae_frames": 1.50704,
1107
+ "val_loss": 27.60484,
1108
+ "val_mel_l1": 0.97122,
1109
+ "val_mr_stft": 2.04703,
1110
+ "val_kl_loss": 3.97215,
1111
+ "val_dur_loss": 0.43763,
1112
+ "val_dur_mae_frames": 1.49866,
1113
+ "per_lang": {
1114
+ "sa": {
1115
+ "n": 36,
1116
+ "val_kl": 4.0612,
1117
+ "val_dur_mae_frames": 1.6878,
1118
+ "val_mel_l1": 1.0091
1119
+ },
1120
+ "hi": {
1121
+ "n": 18,
1122
+ "val_kl": 3.8703,
1123
+ "val_dur_mae_frames": 1.5056,
1124
+ "val_mel_l1": 0.9346
1125
+ },
1126
+ "kn": {
1127
+ "n": 18,
1128
+ "val_kl": 3.9557,
1129
+ "val_dur_mae_frames": 1.5155,
1130
+ "val_mel_l1": 0.9798
1131
+ },
1132
+ "ta": {
1133
+ "n": 18,
1134
+ "val_kl": 3.7681,
1135
+ "val_dur_mae_frames": 1.4534,
1136
+ "val_mel_l1": 0.9486
1137
+ },
1138
+ "te": {
1139
+ "n": 18,
1140
+ "val_kl": 4.0285,
1141
+ "val_dur_mae_frames": 1.3503,
1142
+ "val_mel_l1": 0.9384
1143
+ },
1144
+ "ml": {
1145
+ "n": 18,
1146
+ "val_kl": 4.06,
1147
+ "val_dur_mae_frames": 1.2901,
1148
+ "val_mel_l1": 0.979
1149
+ }
1150
+ }
1151
+ },
1152
+ {
1153
+ "epoch": 20,
1154
+ "step": 2520,
1155
+ "epoch_sec": 20.11,
1156
+ "lr_head": 6.06e-05,
1157
+ "lr_backbone": 1.33e-05,
1158
+ "is_best": true,
1159
+ "train_loss": 26.63825,
1160
+ "train_mel_l1": 0.92621,
1161
+ "train_mr_stft": 2.01087,
1162
+ "train_kl_loss": 3.89764,
1163
+ "train_dur_loss": 0.44149,
1164
+ "train_dur_mae_frames": 1.50608,
1165
+ "val_loss": 27.21235,
1166
+ "val_mel_l1": 0.94692,
1167
+ "val_mr_stft": 2.02553,
1168
+ "val_kl_loss": 4.01842,
1169
+ "val_dur_loss": 0.44071,
1170
+ "val_dur_mae_frames": 1.49733,
1171
+ "per_lang": {
1172
+ "sa": {
1173
+ "n": 36,
1174
+ "val_kl": 4.1527,
1175
+ "val_dur_mae_frames": 1.6842,
1176
+ "val_mel_l1": 0.9604
1177
+ },
1178
+ "hi": {
1179
+ "n": 18,
1180
+ "val_kl": 4.032,
1181
+ "val_dur_mae_frames": 1.5125,
1182
+ "val_mel_l1": 0.9437
1183
+ },
1184
+ "kn": {
1185
+ "n": 18,
1186
+ "val_kl": 3.8209,
1187
+ "val_dur_mae_frames": 1.5132,
1188
+ "val_mel_l1": 0.9323
1189
+ },
1190
+ "ta": {
1191
+ "n": 18,
1192
+ "val_kl": 3.9889,
1193
+ "val_dur_mae_frames": 1.4461,
1194
+ "val_mel_l1": 0.9256
1195
+ },
1196
+ "te": {
1197
+ "n": 18,
1198
+ "val_kl": 3.9578,
1199
+ "val_dur_mae_frames": 1.3456,
1200
+ "val_mel_l1": 0.9344
1201
+ },
1202
+ "ml": {
1203
+ "n": 18,
1204
+ "val_kl": 4.024,
1205
+ "val_dur_mae_frames": 1.2955,
1206
+ "val_mel_l1": 0.9717
1207
+ }
1208
+ }
1209
+ },
1210
+ {
1211
+ "epoch": 21,
1212
+ "step": 2646,
1213
+ "epoch_sec": 19.82,
1214
+ "lr_head": 5.1e-05,
1215
+ "lr_backbone": 1.12e-05,
1216
+ "is_best": true,
1217
+ "train_loss": 26.59179,
1218
+ "train_mel_l1": 0.92518,
1219
+ "train_mr_stft": 2.00416,
1220
+ "train_kl_loss": 3.88807,
1221
+ "train_dur_loss": 0.44005,
1222
+ "train_dur_mae_frames": 1.50244,
1223
+ "val_loss": 26.87383,
1224
+ "val_mel_l1": 0.93228,
1225
+ "val_mr_stft": 2.0204,
1226
+ "val_kl_loss": 3.98295,
1227
+ "val_dur_loss": 0.43507,
1228
+ "val_dur_mae_frames": 1.48678,
1229
+ "per_lang": {
1230
+ "sa": {
1231
+ "n": 36,
1232
+ "val_kl": 4.0941,
1233
+ "val_dur_mae_frames": 1.6749,
1234
+ "val_mel_l1": 0.9447
1235
+ },
1236
+ "hi": {
1237
+ "n": 18,
1238
+ "val_kl": 4.0062,
1239
+ "val_dur_mae_frames": 1.5032,
1240
+ "val_mel_l1": 0.9485
1241
+ },
1242
+ "kn": {
1243
+ "n": 18,
1244
+ "val_kl": 3.9196,
1245
+ "val_dur_mae_frames": 1.5018,
1246
+ "val_mel_l1": 0.9178
1247
+ },
1248
+ "ta": {
1249
+ "n": 18,
1250
+ "val_kl": 3.8499,
1251
+ "val_dur_mae_frames": 1.4425,
1252
+ "val_mel_l1": 0.9217
1253
+ },
1254
+ "te": {
1255
+ "n": 18,
1256
+ "val_kl": 3.926,
1257
+ "val_dur_mae_frames": 1.3312,
1258
+ "val_mel_l1": 0.9185
1259
+ },
1260
+ "ml": {
1261
+ "n": 18,
1262
+ "val_kl": 3.9907,
1263
+ "val_dur_mae_frames": 1.279,
1264
+ "val_mel_l1": 0.9302
1265
+ }
1266
+ }
1267
+ },
1268
+ {
1269
+ "epoch": 22,
1270
+ "step": 2772,
1271
+ "epoch_sec": 19.73,
1272
+ "lr_head": 4.41e-05,
1273
+ "lr_backbone": 9.6e-06,
1274
+ "is_best": false,
1275
+ "train_loss": 26.56703,
1276
+ "train_mel_l1": 0.92297,
1277
+ "train_mr_stft": 2.00174,
1278
+ "train_kl_loss": 3.903,
1279
+ "train_dur_loss": 0.43858,
1280
+ "train_dur_mae_frames": 1.49936,
1281
+ "val_loss": 27.54291,
1282
+ "val_mel_l1": 0.96588,
1283
+ "val_mr_stft": 2.04883,
1284
+ "val_kl_loss": 4.00316,
1285
+ "val_dur_loss": 0.43162,
1286
+ "val_dur_mae_frames": 1.47932,
1287
+ "per_lang": {
1288
+ "sa": {
1289
+ "n": 36,
1290
+ "val_kl": 4.082,
1291
+ "val_dur_mae_frames": 1.6651,
1292
+ "val_mel_l1": 0.9827
1293
+ },
1294
+ "hi": {
1295
+ "n": 18,
1296
+ "val_kl": 3.984,
1297
+ "val_dur_mae_frames": 1.4936,
1298
+ "val_mel_l1": 0.957
1299
+ },
1300
+ "kn": {
1301
+ "n": 18,
1302
+ "val_kl": 3.9403,
1303
+ "val_dur_mae_frames": 1.5012,
1304
+ "val_mel_l1": 0.9738
1305
+ },
1306
+ "ta": {
1307
+ "n": 18,
1308
+ "val_kl": 3.9137,
1309
+ "val_dur_mae_frames": 1.4387,
1310
+ "val_mel_l1": 0.9386
1311
+ },
1312
+ "te": {
1313
+ "n": 18,
1314
+ "val_kl": 3.936,
1315
+ "val_dur_mae_frames": 1.327,
1316
+ "val_mel_l1": 0.9497
1317
+ },
1318
+ "ml": {
1319
+ "n": 18,
1320
+ "val_kl": 4.084,
1321
+ "val_dur_mae_frames": 1.2647,
1322
+ "val_mel_l1": 0.9765
1323
+ }
1324
+ }
1325
+ },
1326
+ {
1327
+ "epoch": 23,
1328
+ "step": 2898,
1329
+ "epoch_sec": 20.26,
1330
+ "lr_head": 3.98e-05,
1331
+ "lr_backbone": 8.7e-06,
1332
+ "is_best": false,
1333
+ "train_loss": 26.49503,
1334
+ "train_mel_l1": 0.92149,
1335
+ "train_mr_stft": 2.00025,
1336
+ "train_kl_loss": 3.87175,
1337
+ "train_dur_loss": 0.4401,
1338
+ "train_dur_mae_frames": 1.50229,
1339
+ "val_loss": 27.4509,
1340
+ "val_mel_l1": 0.96306,
1341
+ "val_mr_stft": 2.03704,
1342
+ "val_kl_loss": 3.9766,
1343
+ "val_dur_loss": 0.43811,
1344
+ "val_dur_mae_frames": 1.49432,
1345
+ "per_lang": {
1346
+ "sa": {
1347
+ "n": 36,
1348
+ "val_kl": 4.0983,
1349
+ "val_dur_mae_frames": 1.6817,
1350
+ "val_mel_l1": 0.9837
1351
+ },
1352
+ "hi": {
1353
+ "n": 18,
1354
+ "val_kl": 3.9687,
1355
+ "val_dur_mae_frames": 1.5048,
1356
+ "val_mel_l1": 0.972
1357
+ },
1358
+ "kn": {
1359
+ "n": 18,
1360
+ "val_kl": 3.894,
1361
+ "val_dur_mae_frames": 1.5157,
1362
+ "val_mel_l1": 0.9718
1363
+ },
1364
+ "ta": {
1365
+ "n": 18,
1366
+ "val_kl": 3.9438,
1367
+ "val_dur_mae_frames": 1.4532,
1368
+ "val_mel_l1": 0.9174
1369
+ },
1370
+ "te": {
1371
+ "n": 18,
1372
+ "val_kl": 3.9235,
1373
+ "val_dur_mae_frames": 1.3383,
1374
+ "val_mel_l1": 0.9469
1375
+ },
1376
+ "ml": {
1377
+ "n": 18,
1378
+ "val_kl": 3.9097,
1379
+ "val_dur_mae_frames": 1.2848,
1380
+ "val_mel_l1": 0.9661
1381
+ }
1382
+ }
1383
+ },
1384
+ {
1385
+ "epoch": 24,
1386
+ "step": 3024,
1387
+ "epoch_sec": 19.46,
1388
+ "lr_head": 3.84e-05,
1389
+ "lr_backbone": 8.4e-06,
1390
+ "is_best": false,
1391
+ "train_loss": 26.43176,
1392
+ "train_mel_l1": 0.91752,
1393
+ "train_mr_stft": 1.99837,
1394
+ "train_kl_loss": 3.88231,
1395
+ "train_dur_loss": 0.43814,
1396
+ "train_dur_mae_frames": 1.50001,
1397
+ "val_loss": 27.2696,
1398
+ "val_mel_l1": 0.9553,
1399
+ "val_mr_stft": 2.03501,
1400
+ "val_kl_loss": 3.95976,
1401
+ "val_dur_loss": 0.4328,
1402
+ "val_dur_mae_frames": 1.47829,
1403
+ "per_lang": {
1404
+ "sa": {
1405
+ "n": 36,
1406
+ "val_kl": 4.0751,
1407
+ "val_dur_mae_frames": 1.6559,
1408
+ "val_mel_l1": 0.9911
1409
+ },
1410
+ "hi": {
1411
+ "n": 18,
1412
+ "val_kl": 3.9222,
1413
+ "val_dur_mae_frames": 1.4992,
1414
+ "val_mel_l1": 0.9513
1415
+ },
1416
+ "kn": {
1417
+ "n": 18,
1418
+ "val_kl": 3.8993,
1419
+ "val_dur_mae_frames": 1.5072,
1420
+ "val_mel_l1": 0.9476
1421
+ },
1422
+ "ta": {
1423
+ "n": 18,
1424
+ "val_kl": 3.9094,
1425
+ "val_dur_mae_frames": 1.428,
1426
+ "val_mel_l1": 0.919
1427
+ },
1428
+ "te": {
1429
+ "n": 18,
1430
+ "val_kl": 3.8974,
1431
+ "val_dur_mae_frames": 1.3318,
1432
+ "val_mel_l1": 0.9436
1433
+ },
1434
+ "ml": {
1435
+ "n": 18,
1436
+ "val_kl": 3.9399,
1437
+ "val_dur_mae_frames": 1.27,
1438
+ "val_mel_l1": 0.9434
1439
+ }
1440
+ }
1441
+ }
1442
+ ]
vocab.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "vocab_tokens": [
3
+ "फ",
4
+ "4",
5
+ "1",
6
+ "-",
7
+ "अ",
8
+ "इ",
9
+ "8",
10
+ " ",
11
+ "0",
12
+ "छ",
13
+ "न",
14
+ "ए",
15
+ "ऐ",
16
+ "़",
17
+ "ष",
18
+ "ि",
19
+ "ँ",
20
+ "च",
21
+ "2",
22
+ "ऑ",
23
+ "थ",
24
+ "भ",
25
+ "ी",
26
+ "‍",
27
+ "ॅ",
28
+ "3",
29
+ "ञ",
30
+ "ै",
31
+ "ु",
32
+ "ठ",
33
+ "ं",
34
+ "ॉ",
35
+ "उ",
36
+ "_",
37
+ "ई",
38
+ "ः",
39
+ "ह",
40
+ "ध",
41
+ "ल",
42
+ "र",
43
+ "स",
44
+ "ब",
45
+ "ख",
46
+ "ण",
47
+ "'",
48
+ "`",
49
+ "व",
50
+ "घ",
51
+ "प",
52
+ "ग",
53
+ "ढ",
54
+ "य",
55
+ "े",
56
+ "्",
57
+ "ा",
58
+ "आ",
59
+ "ड",
60
+ "ज",
61
+ "झ",
62
+ "श",
63
+ "औ",
64
+ "ो",
65
+ "द",
66
+ "ृ",
67
+ "ौ",
68
+ "ऊ",
69
+ "ू",
70
+ "ओ",
71
+ "ट",
72
+ "त",
73
+ "क",
74
+ "म",
75
+ "<unk>",
76
+ "ळ",
77
+ "ऴ",
78
+ "ऱ",
79
+ "ऩ",
80
+ "ऎ",
81
+ "ऒ",
82
+ "ॆ",
83
+ "ॊ",
84
+ "ॠ",
85
+ "ॄ",
86
+ "।",
87
+ "॥"
88
+ ],
89
+ "language_ids": {
90
+ "sa": 0,
91
+ "hi": 1,
92
+ "kn": 2,
93
+ "ta": 3,
94
+ "te": 4,
95
+ "ml": 5
96
+ },
97
+ "sampling_rate": 16000,
98
+ "hop_length": 256
99
+ }