Improve number handling

app.py

@@ -7,7 +7,6 @@ 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
 import torch
 
@@ -46,11 +45,15 @@ if synthesizer is None:
     raise NameError("model not found")
 
 def tts(text: str, stress: str):
-    text = preprocess_text(text)
+    print("============================")
+    print("Original text:", text)
+    print("Stress:", stress)
+    print("Time:", datetime.utcnow())
+    autostress = True if stress == StressOption.AutomaticStress.value else False
+    text = preprocess_text(text, autostress)
     text_limit = 1200
     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, stress, datetime.utcnow())
+    print("Converted:", text)
 
     with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
         with torch.no_grad():
@@ -81,6 +84,7 @@ iface = gr.Interface(
         ["Введ+іть, б+удь л+аска, сво+є р+ечення.", StressOption.ManualStress.value],
         ["Введіть, будь ласка, своє речення.", StressOption.ManualStress.value],
         ["Привіт, як тебе звати?", StressOption.AutomaticStress.value],
+        ["Договір підписано 4 квітня 1949 року.", StressOption.AutomaticStress.value],
     ]
 )
 iface.launch(enable_queue=True, prevent_thread_lock=True)

formatter.py

@@ -1,18 +1,42 @@
-def preprocess_text(text):
+import num2words
+import re
+from stress import sentence_to_stress
+
+def preprocess_text(text, autostress=False):
+    # currencies
+    text = text.replace("$", "долар")
+    text = text.replace("₴", "гривня")
+    text = text.replace("€", "євро")
     # replace apostrophe
     text = text.replace("`", "'")
     text = text.replace("ʼ", "'")
     # numbers
-    text = text.replace("1", "од+ин ")
-    text = text.replace("2", "дв+а ")
-    text = text.replace("3", "тр+и ")
-    text = text.replace("4", "чот+ири ")
-    text = text.replace("5", "п'+ять ")
-    text = text.replace("6", "ш+ість ")
-    text = text.replace("7", "с+ім ")
-    text = text.replace("8", "в+ісім ")
-    text = text.replace("9", "д+ев'ять ")
-    text = text.replace("0", "н+уль ")
+    text = re.sub(r'(\d)\s+(\d)', r'\1\2', text)
+
+    def detect_num_and_convert(word):
+        numbers = "0123456789,."
+        is_number = all(map(lambda x: x in numbers, word))
+        if is_number:
+            try:
+                return num2words.num2words(word, lang="uk")
+            except:
+                return word
+        else:
+            return word
+
+    text = " ".join([detect_num_and_convert(word) for word in text.split(" ")])
+
+    # fallback numbers
+    text = text.replace("1", "один ")
+    text = text.replace("2", "два ")
+    text = text.replace("3", "три ")
+    text = text.replace("4", "чотири ")
+    text = text.replace("5", "п'ять ")
+    text = text.replace("6", "шість ")
+    text = text.replace("7", "сім ")
+    text = text.replace("8", "вісім ")
+    text = text.replace("9", "дев'ять ")
+    text = text.replace("0", "нуль ")
     # speak english alphabet using brute force transliteration
     english = {
         "a": "а",
@@ -47,9 +71,14 @@ def preprocess_text(text):
         text = text.replace(english_char.upper(),  english[english_char].upper())
         text = text.replace(english_char, english[english_char])
 
-    # TODO: autostress support here
+    if autostress:
+        text = sentence_to_stress(text)
+
     return text
 
 
 if __name__ == "__main__":
-    print(preprocess_text("Quality of life update"))
+    print(preprocess_text("Quality of life update"))
+    print(preprocess_text("Він украв 20000000 $"))
+    print(preprocess_text("111 000 000 000 доларів державного боргу."))
+    print(preprocess_text("11100000001 доларів державного боргу."))