Update backend/app.py

backend/app.py

@@ -17,7 +17,6 @@ from gtts.tts import gTTSError
 mms_model = None
 mms_tokenizer = None
 # Define a writable cache directory for Hugging Face models
-os.environ["HOME"] = "/tmp"  # Set HOME to /tmp for writable cache on Spaces
 CACHE_DIR = os.environ.get("TRANSFORMERS_CACHE")
 
 
@@ -104,12 +103,10 @@ def text_to_speech():
             import torch
             import soundfile as sf
 
-            # Import uroman for romanization
-            from uroman import uroman
-
-            # Romanize the text to handle non-Roman characters
-            text = uroman(text)
-            print(f"Romanized text: {text}")
+            # The transformers tokenizer will automatically use uroman if it's installed.
+            # No explicit call is needed.
+            if re.search(r"[^A-Za-z0-9\s\.,\?!;:'\"\-]", text):
+                print("Text contains non-Roman characters. Relying on tokenizer's automatic romanization.")
 
             inputs = mms_tokenizer(text, return_tensors="pt")
             try: