change to ug

app.py

@@ -6,6 +6,8 @@ from fastapi.middleware.cors import CORSMiddleware
 import io
 import soundfile as sf
 from pydantic import BaseModel
+import string
+import unicodedata
 
 
 app = FastAPI()
@@ -19,6 +21,68 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
+def fix_string(batch):
+    batch = batch.lower()
+    batch = unicodedata.normalize('NFKC', batch)
+    ## replace ug chars
+    # Replace 'ژ' with 'ج'
+    batch = batch.replace('ژ', 'ج')
+    batch = batch.replace('ک', 'ك')
+    batch = batch.replace('ی', 'ى')
+    # batch = batch.replace('ه', 'ە')
+    batch = batch.replace('ه', 'ە')
+    ## replace nums
+    numbers_to_uyghur_map = {
+        '0': ' نۆل ',
+        '1': ' بىر ',
+        '2': ' ئىككى ',
+        '3': ' ئۈچ ',
+        '4': ' تۆت ',
+        '5': ' بەش ',
+        '6': ' ئالتە ',
+        '7': ' يەتتە ',
+        '8': ' سەككىز ',
+        '9': ' توققۇز '
+    }
+    for num_char, uyghur_char in numbers_to_uyghur_map.items():
+        batch = batch.replace(num_char, uyghur_char)
+    ## replace en chars
+    english_to_uyghur_map = {
+        'a': ' ئېي ',
+        'b': ' بى ',
+        'c': ' سى ',
+        'd': ' دى ',
+        'e': ' ئى ',
+        'f': ' ئەف ',
+        'g': ' جى ',
+        'h': ' ئېچ ',
+        'i': ' ئاي ',
+        'j': ' جېي ',
+        'k': ' کېي ',
+        'l': ' ئەل ',
+        'm': ' ئەم ',
+        'n': ' ئېن ',
+        'o': ' ئو ',
+        'p': ' پى ',
+        'q': ' كىيۇ ',
+        'r': ' ئار ',
+        's': ' ئەس ',
+        't': ' تى ',
+        'u': ' يۇ ',
+        'v': ' ۋى ',
+        'w': ' دابىلىيۇ ',
+        'x': ' ئېكىس ',
+        'y': ' ۋاي ',
+        'z': ' زى ',
+    }
+    for eng_char, uyghur_char in english_to_uyghur_map.items():
+        batch = batch.replace(eng_char, uyghur_char)
+    # batch = batch.replace('e', ' ئې ')
+    # Optional: Collapse multiple spaces into one
+    batch = ' '.join(batch.split())
+    return batch
+
+
 model = VitsModel.from_pretrained("facebook/mms-tts-uig-script_arabic")
 tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-uig-script_arabic")
 
@@ -36,7 +100,7 @@ curl -X POST https://piyazon-tts-piyazon.hf.space/generate-tts \
 async def generate_tts(input: TextInput):
     try:
         # Tokenize input text
-        inputs = tokenizer(input.text, return_tensors="pt")
+        inputs = tokenizer(fix_string(input.text), return_tensors="pt")
         
         # Generate waveform
         with torch.no_grad():