Add improved "Mykyta" model with stress support

.gitignore

@@ -129,4 +129,5 @@ dmypy.json
 .pyre/
 
 # model files
-*.pth.tar
+*.pth.tar
+*.pth

app.py

@@ -9,7 +9,7 @@ from os.path import exists
 from formatter import preprocess_text
 
 MODEL_NAMES = [
-    "uk/mai/vits-tts"
+    "uk/mykyta/vits-tts"
 ]
 MODELS = {}
 
@@ -28,11 +28,11 @@ def download(url, file_name):
 
 for MODEL_NAME in MODEL_NAMES:
     print(f"downloading {MODEL_NAME}")
-    release_number = "1.0.0"
-    model_link = f"https://github.com/robinhad/ukrainian-tts/releases/download/v{release_number}/model.pth.tar"
-    config_link = f"https://github.com/robinhad/ukrainian-tts/releases/download/v{release_number}/config.json"
+    release_number = "v2.0.0-beta"
+    model_link = f"https://github.com/robinhad/ukrainian-tts/releases/download/{release_number}/model.pth"
+    config_link = f"https://github.com/robinhad/ukrainian-tts/releases/download/{release_number}/config.json"
 
-    model_path = "model.pth.tar"
+    model_path = "model.pth"
     config_path = "config.json"
 
     download(model_link, model_path)
@@ -74,7 +74,7 @@ iface = gr.Interface(
     outputs=gr.outputs.Audio(label="Output"),
     title="🐸💬🇺🇦 - Coqui TTS",
     theme="huggingface",
-    description="Україномовний🇺🇦 TTS за допомогою Coqui TTS (для апострофа використовуйте символ у трикутних дужках: <'>)",
+    description="Україномовний🇺🇦 TTS за допомогою Coqui TTS (для наголосу використовуйте + перед голосною)",
     article="Якщо вам подобається, підтримайте за посиланням: [SUPPORT LINK](https://send.monobank.ua/jar/48iHq4xAXm),  " +
     "Github: [https://github.com/robinhad/ukrainian-tts](https://github.com/robinhad/ukrainian-tts)",
 )

config.json

@@ -1,32 +1,56 @@
 {
-    "model": "vits",
-    "run_name": "vits_ljspeech",
-    "run_description": "",
-    "epochs": 1000,
-    "batch_size": 18,
-    "eval_batch_size": 16,
-    "mixed_precision": true,
-    "scheduler_after_epoch": true,
-    "run_eval": true,
-    "test_delay_epochs": -1,
-    "print_eval": true,
-    "dashboard_logger": "tensorboard",
+    "output_path": "./model",
+    "logger_uri": null,
+    "run_name": "vits_mykyta",
+    "project_name": null,
+    "run_description": "\ud83d\udc38Coqui trainer run.",
     "print_step": 25,
     "plot_step": 100,
     "model_param_stats": false,
-    "project_name": null,
-    "log_model_step": null,
     "wandb_entity": null,
+    "dashboard_logger": "tensorboard",
+    "log_model_step": null,
     "save_step": 10000,
-    "checkpoint": true,
-    "keep_all_best": false,
-    "keep_after": 10000,
-    "num_loader_workers": 12,
-    "num_eval_loader_workers": 12,
-    "use_noise_augment": false,
-    "output_path": "./ukrainian-vits",
+    "save_n_checkpoints": 5,
+    "save_checkpoints": true,
+    "save_all_best": false,
+    "save_best_after": 10000,
+    "target_loss": null,
+    "print_eval": false,
+    "test_delay_epochs": -1,
+    "run_eval": true,
     "distributed_backend": "nccl",
     "distributed_url": "tcp://localhost:54321",
+    "mixed_precision": true,
+    "epochs": 1000,
+    "batch_size": 32,
+    "eval_batch_size": 16,
+    "grad_clip": [
+        1000.0,
+        1000.0
+    ],
+    "scheduler_after_epoch": true,
+    "lr": 0.001,
+    "optimizer": "AdamW",
+    "optimizer_params": {
+        "betas": [
+            0.8,
+            0.99
+        ],
+        "eps": 1e-09,
+        "weight_decay": 0.01
+    },
+    "lr_scheduler": "",
+    "lr_scheduler_params": {},
+    "use_grad_scaler": false,
+    "cudnn_enable": true,
+    "cudnn_deterministic": false,
+    "cudnn_benchmark": true,
+    "training_seed": 54321,
+    "model": "vits",
+    "num_loader_workers": 8,
+    "num_eval_loader_workers": 8,
+    "use_noise_augment": false,
     "audio": {
         "fft_size": 1024,
         "win_length": 1024,
@@ -34,23 +58,27 @@
         "frame_shift_ms": null,
         "frame_length_ms": null,
         "stft_pad_mode": "reflect",
-        "sample_rate": 16000,
+        "sample_rate": 48000,
         "resample": false,
         "preemphasis": 0.0,
         "ref_level_db": 20,
-        "do_sound_norm": false,
-        "log_func": "np.log",
+        "do_sound_norm": true,
+        "log_func": "np.log10",
         "do_trim_silence": true,
         "trim_db": 45,
-        "power": 1.3,
+        "do_rms_norm": false,
+        "db_level": null,
+        "power": 1.5,
         "griffin_lim_iters": 60,
         "num_mels": 80,
         "mel_fmin": 0.0,
         "mel_fmax": null,
         "spec_gain": 1,
-        "do_amp_to_db_linear": false,
+        "do_amp_to_db_linear": true,
         "do_amp_to_db_mel": true,
-        "signal_norm": false,
+        "pitch_fmax": 640.0,
+        "pitch_fmin": 0.0,
+        "signal_norm": true,
         "min_level_db": -100,
         "symmetric_norm": true,
         "max_norm": 4.0,
@@ -58,68 +86,73 @@
         "stats_path": null
     },
     "use_phonemes": false,
-    "use_espeak_phonemes": true,
-    "phoneme_language": null,
-    "compute_input_seq_cache": true,
+    "phonemizer": null,
+    "phoneme_language": "uk",
+    "compute_input_seq_cache": false,
     "text_cleaner": "basic_cleaners",
     "enable_eos_bos_chars": false,
     "test_sentences_file": "",
-    "phoneme_cache_path": "./ukrainian/phoneme_cache",
+    "phoneme_cache_path": "./phoneme_cache",
     "characters": {
-        "pad": "_",
-        "eos": "~",
-        "bos": "^",
-        "characters": "!',-.:;?\u0410\u0411\u0412\u0413\u0490\u0414\u0415\u0404\u0416\u0417\u0418\u0406\u0407\u0419\u041a\u041b\u041c\u041d\u041e\u041f\u0420\u0421\u0422\u0423\u0424\u0425\u0426\u0427\u0428\u0429\u042c\u042e\u042f\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 ",
-        "punctuations": "!',-.:;? ",
+        "characters_class": "TTS.tts.models.vits.VitsCharacters",
+        "vocab_dict": null,
+        "pad": "<PAD>",
+        "eos": "<EOS>",
+        "bos": "<BOS>",
+        "blank": "<BLNK>",
+        "characters": "!\"'(),-/:;?\u00ab\u00bb+\u0430\u0431\u0432\u0433\u0434\u0435\u0436\u0437\u0438\u0439\u043a\u043b\u043c\u043d\u043e\u043f\u0440\u0441\u0442\u0443\u0444\u0445\u0446\u0447\u0448\u0449\u044c\u044e\u044f\u0454\u0456\u0457\u0491\u2013\u2014\u2015\u2018\u2019\u201c\u201d\u201e\u2026 ",
+        "punctuations": "!\"'(),-/:;?\u00ab\u00bb+\u2013\u2014\u2015\u2018\u2019\u201c\u201d\u201e\u2026 ",
         "phonemes": null,
-        "unique": true
+        "is_unique": true,
+        "is_sorted": true
     },
-    "batch_group_size": 5,
+    "add_blank": true,
+    "batch_group_size": 0,
     "loss_masking": null,
-    "sort_by_audio_len": true,
-    "min_seq_len": 0,
-    "max_seq_len": 500000,
+    "sort_by_audio_len": false,
+    "min_audio_len": 24000,
+    "max_audio_len": 624000,
+    "min_text_len": 1,
+    "max_text_len": Infinity,
     "compute_f0": false,
     "compute_linear_spec": true,
-    "add_blank": true,
+    "precompute_num_workers": 8,
+    "start_by_longest": false,
     "datasets": [
         {
             "name": "ljspeech",
-            "path": "./Data/uk_UK/by_book/female/sumska/kaydasheva",
-            "meta_file_train": "metadata.csv",
-            "ununsed_speakers": null,
-            "meta_file_val": "",
-            "meta_file_attn_mask": ""
-        },
-        {
-            "name": "ljspeech",
-            "path": "./Data/uk_UK/by_book/female/sumska/mykola_djerya",
+            "path": "./dataset",
             "meta_file_train": "metadata.csv",
-            "ununsed_speakers": null,
+            "ignored_speakers": null,
+            "language": "",
             "meta_file_val": "",
             "meta_file_attn_mask": ""
         }
     ],
-    "optimizer": "AdamW",
-    "optimizer_params": {
-        "betas": [
-            0.8,
-            0.99
-        ],
-        "eps": 1e-09,
-        "weight_decay": 0.01
-    },
-    "lr_scheduler": "",
-    "lr_scheduler_params": {},
     "test_sentences": [
-        "\u0413\u043e\u0432\u043e\u0440\u0438 \u043d\u0456\u0431\u0438 \u0442\u0438 \u0436\u0438\u0432\u0438\u0439!",
-        "\u041f\u043e\u043b \u043f\u0435\u0440\u0435\u0442\u043d\u0443\u0432 \u043f\u0443\u0441\u0442\u0435\u043b\u044e",
-        "\u041f\u0440\u0438\u0432\u0456\u0442, \u0441\u0432\u0456\u0442\u0435!"
+        [
+            "\u0413\u043e\u0432\u043e\u0440+\u0438, \u043d+\u0456\u0431\u0438 \u0442+\u0438 \u0436\u0438\u0432+\u0438\u0439!"
+        ],
+        [
+            "\u041f\u0435\u0440\u0435\u043f\u0440+\u043e\u0448\u0443\u044e, \u0414+\u0435\u0439\u0432\u0435, \u043d+\u0430 \u0436+\u0430\u043b\u044c, +\u044f \u043d+\u0435 \u043c+\u043e\u0436\u0443 \u0437\u0440\u043e\u0431+\u0438\u0442\u0438 \u0446+\u0435."
+        ],
+        [
+            "\u0425\u0442+\u043e \u0442+\u0438 \u0442\u0430\u043a+\u0438\u0439 +\u0456 +\u044f\u043a \u0442\u0435\u0431+\u0435 \u0437\u0432+\u0430\u0442\u0438?"
+        ],
+        [
+            "\u041a+\u0430\u043c'\u044f\u043d+\u0435\u0446\u044c-\u041f\u043e\u0434+\u0456\u043b\u044c\u0441\u044c\u043a\u0438\u0439 - \u043c+\u0456\u0441\u0442\u043e \u0432 \u0425\u043c\u0435\u043b\u044c\u043d+\u0438\u0446\u044c\u043a\u0456\u0439 +\u043e\u0431\u043b\u0430\u0441\u0442\u0456 \u0423\u043a\u0440\u0430+\u0457\u043d\u0438, \u0446+\u0435\u043d\u0442\u0440 \u041a\u0430\u043c'\u044f\u043d+\u0435\u0446\u044c-\u041f\u043e\u0434+\u0456\u043b\u044c\u0441\u044c\u043a\u043e\u0457 \u043c\u0456\u0441\u044c\u043a+\u043e\u0457 \u043e\u0431'+\u0454\u0434\u043d\u0430\u043d\u043e\u0457 \u0442\u0435\u0440\u0438\u0442\u043e\u0440\u0456+\u0430\u043b\u044c\u043d\u043e\u0457 \u0433\u0440\u043e\u043c+\u0430\u0434\u0438 +\u0456 \u041a\u0430\u043c'\u044f\u043d+\u0435\u0446\u044c-\u041f\u043e\u0434+\u0456\u043b\u044c\u0441\u044c\u043a\u043e\u0433\u043e \u0440\u0430\u0439+\u043e\u043d\u0443."
+        ]
     ],
+    "eval_split_max_size": null,
+    "eval_split_size": 0.01,
+    "use_speaker_weighted_sampler": false,
+    "speaker_weighted_sampler_alpha": 1.0,
+    "use_language_weighted_sampler": false,
+    "language_weighted_sampler_alpha": 1.0,
     "model_args": {
-        "num_chars": 86,
+        "num_chars": 83,
         "out_channels": 513,
-        "spec_segment_size": 24,
+        "spec_segment_size": 32,
         "hidden_channels": 192,
         "hidden_channels_ffn_text_encoder": 768,
         "num_heads_text_encoder": 2,
@@ -181,16 +214,25 @@
         "use_speaker_embedding": false,
         "num_speakers": 0,
         "speakers_file": null,
+        "d_vector_file": null,
         "speaker_embedding_channels": 256,
         "use_d_vector_file": false,
-        "d_vector_file": null,
         "d_vector_dim": 0,
-        "detach_dp_input": true
+        "detach_dp_input": true,
+        "use_language_embedding": false,
+        "embedded_language_dim": 4,
+        "num_languages": 0,
+        "language_ids_file": null,
+        "use_speaker_encoder_as_loss": false,
+        "speaker_encoder_config_path": "",
+        "speaker_encoder_model_path": "",
+        "condition_dp_on_speaker": true,
+        "freeze_encoder": false,
+        "freeze_DP": false,
+        "freeze_PE": false,
+        "freeze_flow_decoder": false,
+        "freeze_waveform_decoder": false
     },
-    "grad_clip": [
-        1000.0,
-        1000.0
-    ],
     "lr_gen": 0.0002,
     "lr_disc": 0.0002,
     "lr_scheduler_gen": "ExponentialLR",
@@ -209,13 +251,18 @@
     "feat_loss_alpha": 1.0,
     "mel_loss_alpha": 45.0,
     "dur_loss_alpha": 1.0,
+    "speaker_encoder_loss_alpha": 1.0,
     "return_wav": true,
     "r": 1,
     "num_speakers": 0,
     "use_speaker_embedding": false,
     "speakers_file": null,
     "speaker_embedding_channels": 256,
+    "language_ids_file": null,
+    "use_language_embedding": false,
     "use_d_vector_file": false,
     "d_vector_file": null,
-    "d_vector_dim": 0
+    "d_vector_dim": 0,
+    "restore_path": "./vits_mykyta-May-06-2022_12+17AM-c410bc58/best_model_90901.pth",
+    "github_branch": "* (HEAD detached at v0.6.2)"
 }

requirements.txt

@@ -1,2 +1,2 @@
-TTS==0.4.1
+TTS==0.6.2
 gradio