Split to sentances for verbalization.

app.py

@@ -20,7 +20,7 @@ prompts_dir = 'voices'
 verbalizer = Verbalizer()
 
 
-def split_to_parts(text):
+def split_to_parts(text, group=True):
     text = re.sub(r'([^.,!:?\-])\n', r'\1. ', text)
     text = text.replace('\n', ' ')
     split_symbols = '.?!:'
@@ -29,7 +29,9 @@ def split_to_parts(text):
     last = len(text)-1
     for i, s in enumerate(text):
         parts[index] += s
-        if s in split_symbols and i < last and text[i+1] == ' ' and len(parts[index]) > 20:
+        if s in split_symbols and i < last and text[i+1] == ' ':
+            if group and len(parts[index]) <= 20:
+                continue
             index += 1
             parts.append('')
     return parts
@@ -64,11 +66,11 @@ models = {
 
 
 def verbalize(text):
-    parts = split_to_parts(text)
+    parts = split_to_parts(text, group=False)
     verbalized = ''
     for part in parts:
         if part.strip():
-            verbalized += verbalizer.process_text(part)[0] + ' '
+            verbalized += verbalizer.process_text(part.strip())[0] + ' '
     return verbalized
 
 description = f'''

verbalizer.py

@@ -22,7 +22,7 @@ class Verbalizer():
         self.translator = ctranslate2.Translator(
             local_model_path,
             device='cpu',
-            compute_type="int8",  # Use INT8 quantization for CPU
+            compute_type="int8",
         )
         
         # Load tokenizer