Add automatic stress support

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "ukrainian-accentor"]
+	path = ukrainian-accentor
+	url = https://github.com/egorsmkv/ukrainian-accentor.git

app.py

@@ -8,6 +8,12 @@ import requests
 from os.path import exists
 from formatter import preprocess_text
 from datetime import datetime
+from stress import sentence_to_stress
+from enum import Enum
+
+class StressOption(Enum):
+    ManualStress = "Наголоси вручну"
+    AutomaticStress = "Автоматичні наголоси (Beta)"
 
 MODEL_NAMES = [
     "uk/mykyta/vits-tts"
@@ -30,7 +36,7 @@ def download(url, file_name):
 for MODEL_NAME in MODEL_NAMES:
     print(f"downloading {MODEL_NAME}")
     release_number = "v2.0.0-beta"
-    model_link = f"https://github.com/robinhad/ukrainian-tts/releases/download/{release_number}/model.pth"
+    model_link = f"https://github.com/robinhad/ukrainian-tts/releases/download/{release_number}/model-inference.pth"
     config_link = f"https://github.com/robinhad/ukrainian-tts/releases/download/{release_number}/config.json"
 
     model_path = "model.pth"
@@ -43,13 +49,14 @@ for MODEL_NAME in MODEL_NAMES:
     #MODELS[MODEL_NAME] = synthesizer
 
 
-def tts(text: str):
+def tts(text: str, stress: str):
     synthesizer = Synthesizer(
         model_path, config_path, None, None, None,
     )
     text = preprocess_text(text)
-    text_limit = 100
+    text_limit = 150
     text = text if len(text) < text_limit else text[0:text_limit] # mitigate crashes on hf space
+    text = sentence_to_stress(text) if stress == StressOption.AutomaticStress.value else text
     print(text, datetime.utcnow())
     if synthesizer is None:
         raise NameError("model not found")
@@ -68,10 +75,10 @@ iface = gr.Interface(
             label="Input",
             default="Введ+іть, б+удь л+аска, сво+є р+ечення.",
         ),
-        #gr.inputs.Radio(
-        #    label="Виберіть TTS модель",
-        #    choices=MODEL_NAMES,
-        #),
+        gr.inputs.Radio(
+            label="Опції",
+            choices=[option.value for option in StressOption],
+        ),
     ],
     outputs=gr.outputs.Audio(label="Output"),
     title="🐸💬🇺🇦 - Coqui TTS",
@@ -79,5 +86,9 @@ iface = gr.Interface(
     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)",
+    examples=[
+        ["Введ+іть, б+удь л+аска, сво+є р+ечення.", StressOption.ManualStress.value],
+        ["Привіт, як тебе звати?", StressOption.AutomaticStress.value]
+    ]
 )
-iface.launch(enable_queue=True)
+iface.launch(enable_queue=True, prevent_thread_lock=True)

stress.py

@@ -0,0 +1,60 @@
+from unittest import skip
+from gruut import sentences
+import torch
+
+importer = torch.package.PackageImporter("ukrainian-accentor/accentor-lite.pt")
+accentor = importer.load_pickle("uk-accentor", "model")
+replace_accents = importer.load_pickle("uk-accentor", "replace_accents")
+
+# Using GPU
+# accentor.cuda()
+# Back to CPU
+# accentor.cpu()
+
+alphabet = "абгґдеєжзиіїйклмнопрстуфхцчшщьюя"
+
+def accent_word(word):
+    stressed_words = accentor.predict([word], mode='stress')
+    plused_words = [replace_accents(x) for x in stressed_words]
+    return plused_words[0]
+
+def sentence_to_stress(sentence):
+    words = sentence.split()
+    words = sum([[word, " "] for word in words], start=[])
+    new_list = []
+    for word in words:
+        first_word_sep = list(map(lambda letter: letter in alphabet, word.lower()))
+        if all(first_word_sep):
+            new_list.append(word)
+        else:
+            current_index = 0
+            past_index = 0
+            for letter in first_word_sep:
+                if letter == False:
+                    new_list.append(word[past_index:current_index])
+                    new_list.append(word[current_index])
+                    past_index = current_index + 1
+                current_index += 1
+            new_list.append(word[past_index:current_index])
+    #print(list(filter(lambda x: len(x) > 0, new_list)))
+    for word_index in range(0, len(new_list)):
+        element = new_list[word_index]
+        first_word_sep = list(map(lambda letter: letter in alphabet, element.lower()))
+        if not all(first_word_sep) or len(element) == 0:
+            continue
+        else:
+            new_list[word_index] = accent_word(new_list[word_index])
+
+    return "".join(new_list)
+
+
+if __name__ == "__main__":
+    sentence = "Кам'янець-Подільський - місто в Хмельницькій області України, центр Кам'янець-Подільської міської об'єднаної територіальної громади і Кам'янець-Подільського району."
+    print(sentence_to_stress(sentence))
+    #test_words1 = ["словотворення", "архаїчний", "програма", "а-ля-фуршет"]
+
+    stressed_words = accentor.predict(["привіт"], mode='stress')
+    plused_words = [replace_accents(x) for x in stressed_words]
+
+    print('With stress:', stressed_words)
+    print('With pluses:', plused_words)

ukrainian-accentor

@@ -0,0 +1 @@
+Subproject commit a3dd2cf9341db200853cfd19df142224a47749b2