Updated hifi-V3 config

config.json

@@ -23,7 +23,7 @@
     "distributed_url": "tcp://localhost:54321",
     "mixed_precision": true,
     "epochs": 1000,
-    "batch_size": 16,
+    "batch_size": 32,
     "eval_batch_size": 4,
     "grad_clip": [
         1000,
@@ -45,7 +45,7 @@
     "use_grad_scaler": false,
     "cudnn_enable": true,
     "cudnn_deterministic": false,
-    "cudnn_benchmark": true,
+    "cudnn_benchmark": false,
     "training_seed": 54321,
     "model": "vits",
     "num_loader_workers": 8,
@@ -58,7 +58,7 @@
         "frame_shift_ms": null,
         "frame_length_ms": null,
         "stft_pad_mode": "reflect",
-        "sample_rate": 16000,
+        "sample_rate": 22050,
         "resample": false,
         "preemphasis": 0.0,
         "ref_level_db": 20,
@@ -78,7 +78,7 @@
         "do_amp_to_db_mel": true,
         "pitch_fmax": 640.0,
         "pitch_fmin": 0.0,
-        "signal_norm": false,
+        "signal_norm": true,
         "min_level_db": -100,
         "symmetric_norm": true,
         "max_norm": 4.0,
@@ -100,7 +100,7 @@
         "eos": "<EOS>",
         "bos": "<BOS>",
         "blank": "<BLNK>",
-        "characters": "\u0430\u0431\u0432\u0433\u0491\u0434\u0435\u0454\u0436\u0437\u0438\u0456\u0457\u0439\u043a\u043b\u043c\u043d\u043e\u043f\u0440\u0441\u0442\u0443\u0444\u0445\u0446\u0447\u0448\u0449\u044c\u044e\u044f\u044d\u0451\u044b\u044a",
+        "characters": "\u0430\u0431\u0432\u0433\u0491\u0434\u0435\u0454\u0436\u0437\u0438\u0456\u0457\u0439\u043a\u043b\u043c\u043d\u043e\u043f\u0440\u0441\u0442\u0443\u0444\u0445\u0446\u0447\u0448\u0449\u044c\u044e\u044f\u044d\u0451\u044b\u044a+",
         "punctuations": "!'(),-.:;? ",
         "phonemes": null,
         "is_unique": true,
@@ -111,7 +111,7 @@
     "loss_masking": null,
     "sort_by_audio_len": false,
     "min_audio_len": 32768,
-    "max_audio_len": 224000,
+    "max_audio_len": 132300,
     "min_text_len": 1,
     "max_text_len": Infinity,
     "compute_f0": false,
@@ -120,13 +120,10 @@
     "start_by_longest": false,
     "datasets": [
         {
-            "name": "mailabs",
-            "path": "./logs/uk_UK",
-            "meta_file_train": "",
-            "ignored_speakers": [
-                "obruchov",
-                "shepel"
-            ],
+            "name": "ljspeech",
+            "path": "./datasets/uk_Mykyta",
+            "meta_file_train": "metadata_not.csv",
+            "ignored_speakers": null,
             "language": "uk",
             "meta_file_val": "",
             "meta_file_attn_mask": ""
@@ -135,7 +132,7 @@
     "test_sentences": [
         [
             "\u0412\u0435\u0441\u0435\u043b\u043a\u0430, \u0442\u0430\u043a\u043e\u0436 \u0440\u0430\u0439\u0434\u0443\u0433\u0430 \u043e\u043f\u0442\u0438\u0447\u043d\u0435 \u044f\u0432\u0438\u0449\u0435 \u0432 \u0430\u0442\u043c\u043e\u0441\u0444\u0435\u0440\u0456, \u0449\u043e \u044f\u0432\u043b\u044f\u0454 \u0441\u043e\u0431\u043e\u044e \u043e\u0434\u043d\u0443, \u0434\u0432\u0456 \u0447\u0438 \u0434\u0435\u043a\u0456\u043b\u044c\u043a\u0430 \u0440\u0456\u0437\u043d\u043e\u043a\u043e\u043b\u044c\u043e\u0440\u043e\u0432\u0438\u0445 \u0434\u0443\u0433.",
-            "sumska",
+            "ljspeech",
             null,
             "uk"
         ]
@@ -147,7 +144,7 @@
     "use_language_weighted_sampler": true,
     "language_weighted_sampler_alpha": 1.0,
     "model_args": {
-        "num_chars": 52,
+        "num_chars": 53,
         "out_channels": 513,
         "spec_segment_size": 32,
         "hidden_channels": 192,
@@ -166,38 +163,33 @@
         "resblock_type_decoder": "2",
         "resblock_kernel_sizes_decoder": [
             3,
-            7,
-            11
+            5,
+            7
         ],
         "resblock_dilation_sizes_decoder": [
             [
                 1,
-                3,
-                5
+                2
             ],
             [
-                1,
-                3,
-                5
+                2,
+                6
             ],
             [
-                1,
                 3,
-                5
+                12
             ]
         ],
         "upsample_rates_decoder": [
             8,
             8,
-            2,
-            2
+            4
         ],
-        "upsample_initial_channel_decoder": 512,
+        "upsample_initial_channel_decoder": 256,
         "upsample_kernel_sizes_decoder": [
             16,
             16,
-            4,
-            4
+            8
         ],
         "use_sdp": true,
         "noise_scale": 1.0,

speaker_ids.json

@@ -1,4 +1,4 @@
 {
     "uk": 0,
-    "sumska": 0
+    "mykyta": 0
 }