app

app.py

@@ -1,12 +1,8 @@
 import gradio as gr
 import util
+import tts
 
 # Functions
-def generate_example_pronunciation(input_text, script):
-    # Placeholder for generating example pronunciation
-    example_audio = None  # Replace with actual example audio generation logic
-    return example_audio
-
 def check_pronunciation(input_text, script, user_audio):
     # Placeholder logic for pronunciation checking
     transcript_ugArab_box = "Automatic transcription of your audio (Arabic)..."
@@ -105,7 +101,7 @@ with gr.Blocks() as app:
     )
     
     tts_btn.click(
-        generate_example_pronunciation,
+        tts.generate_example_pronunciation,
         inputs=[input_text, script_choice],
         outputs=[example_audio]
     )

tts.py

@@ -0,0 +1,26 @@
+from transformers import VitsModel, AutoTokenizer
+import torch
+from umsc import UgMultiScriptConverter
+import scipy.io.wavfile
+import os
+
+tts_tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-uig-script_arabic")
+tts_model = VitsModel.from_pretrained("facebook/mms-tts-uig-script_arabic")
+
+def generate_example_pronunciation(input_text, script):
+    # Convert text to uyghur_arabic
+    ug_latn_to_arab = UgMultiScriptConverter('ULS', 'UAS')
+    if not script == "Uyghur Arabic":
+        input_text = ug_latn_to_arab(input_text)
+
+    tts_inputs = tts_tokenizer(input_text, return_tensors="pt")
+    with torch.no_grad():
+        tts_output = tts_model(**tts_inputs).waveform
+
+    # Save to a temporary file
+    output_path = "tts_output.wav"
+    sample_rate = 16000
+    scipy.io.wavfile.write(output_path, rate=sample_rate, data=tts_output.numpy()[0])
+
+    # Return the audio file path
+    return output_path