Update app.py

app.py

@@ -4,6 +4,7 @@ import soundfile as sf
 import spaces
 import os
 import numpy as np
+import re
 from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
 from speechbrain.pretrained import EncoderClassifier
 from datasets import load_dataset
@@ -44,22 +45,94 @@ def prepare_default_embedding(example):
 
 default_embedding = prepare_default_embedding(default_example)
 
+replacements = [
+    ("â", "a"),  # Long a
+    ("ç", "ch"),  # Ch as in "chair"
+    ("ğ", "gh"),  # Silent g or slight elongation of the preceding vowel
+    ("ı", "i"),   # Dotless i
+    ("î", "i"),   # Long i
+    ("ö", "oe"),  # Similar to German ö
+    ("ş", "sh"),  # Sh as in "shoe"
+    ("ü", "ue"),  # Similar to German ü
+    ("û", "u"),   # Long u
+]
+
+number_words = {
+    0: "sıfır", 1: "bir", 2: "iki", 3: "üç", 4: "dört", 5: "beş", 6: "altı", 7: "yedi", 8: "sekiz", 9: "dokuz",
+    10: "on", 11: "on bir", 12: "on iki", 13: "on üç", 14: "on dört", 15: "on beş", 16: "on altı", 17: "on yedi",
+    18: "on sekiz", 19: "on dokuz", 20: "yirmi", 30: "otuz", 40: "kırk", 50: "elli", 60: "altmış", 70: "yetmiş",
+    80: "seksen", 90: "doksan", 100: "yüz", 1000: "bin"
+}
+
+def number_to_words(number):
+    if number < 20:
+        return number_words[number]
+    elif number < 100:
+        tens, unit = divmod(number, 10)
+        return number_words[tens * 10] + (" " + number_words[unit] if unit else "")
+    elif number < 1000:
+        hundreds, remainder = divmod(number, 100)
+        return (number_words[hundreds] + " yüz" if hundreds > 1 else "yüz") + (" " + number_to_words(remainder) if remainder else "")
+    elif number < 1000000:
+        thousands, remainder = divmod(number, 1000)
+        return (number_to_words(thousands) + " bin" if thousands > 1 else "bin") + (" " + number_to_words(remainder) if remainder else "")
+    elif number < 1000000000:
+        millions, remainder = divmod(number, 1000000)
+        return number_to_words(millions) + " milyon" + (" " + number_to_words(remainder) if remainder else "")
+    elif number < 1000000000000:
+        billions, remainder = divmod(number, 1000000000)
+        return number_to_words(billions) + " milyar" + (" " + number_to_words(remainder) if remainder else "")
+    else:
+        return str(number)
+
+def replace_numbers_with_words(text):
+    def replace(match):
+        number = int(match.group())
+        return number_to_words(number)
+
+    # Find the numbers and change with words.
+    result = re.sub(r'\b\d+\b', replace, text)
+
+    return result
+
+def normalize_text(text):
+    # Convert to lowercase
+    text = text.lower()
+    
+    # Replace numbers with words
+    text = replace_numbers_with_words(text)
+    
+    # Apply character replacements
+    for old, new in replacements:
+        text = text.replace(old, new)
+    
+    # Remove punctuation
+    text = re.sub(r'[^\w\s]', '', text)
+    
+    return text
+
 @spaces.GPU(duration = 60)
 def text_to_speech(text, audio_file=None):
-    inputs = processor(text=text, return_tensors="pt").to(device)
+    # Normalize the input text
+    normalized_text = normalize_text(text)
+    
+    inputs = processor(text=normalized_text, return_tensors="pt").to(device)
     
     speaker_embeddings = default_embedding
     
     speech = model.generate_speech(inputs["input_ids"], speaker_embeddings.unsqueeze(0), vocoder=vocoder)
     sf.write("output.wav", speech.cpu().numpy(), samplerate=16000)
-    return "output.wav"
+    return "output.wav", normalized_text
 
 iface = gr.Interface(
     fn=text_to_speech,
     inputs=[
         gr.Textbox(label="Enter Turkish text to convert to speech")
     ],
-    outputs=gr.Audio(label="Generated Speech"),
+    outputs=[
+        gr.Audio(label="Generated Speech"),
+        gr.Textbox(label="Normalized Text")
+    ],
     title="Turkish SpeechT5 Text-to-Speech Demo with Optional Custom Voice",
     description="Enter Turkish text, optionally upload a short audio sample of the target speaker, and listen to the generated speech using the fine-tuned SpeechT5 model."
 )