Update app.py

app.py

@@ -173,15 +173,20 @@ def synthesize_speech(synth_text, target_lang, gender="Male", pace="normal", out
     with open(ref_text_path, encoding='utf-8') as f:
         ref_text = f.read()
 
-    client = Client("ai4bharat/IndicF5")
-    result = client.predict(
-            text=synth_text,
-            ref_audio=handle_file(ref_audio_path),
-            ref_text=ref_text,
-            api_name="/synthesize_speech"
-    )
-    print("synthesize_speech saved to ", result)
-    return result
+    print("> Loading IndicF5 TTS model (ai4bharat/IndicF5)...")
+    indicf5_repo_id = "ai4bharat/IndicF5"
+    tts_model = AutoModel.from_pretrained(indicf5_repo_id, trust_remote_code=True).to(device)
+    
+    audio = tts_model(synth_text, ref_audio_path=ref_audio_path, ref_text=ref_text)
+
+    if audio.dtype == np.int16:
+        audio = audio.astype(np.float32) / 32768.0
+    
+    sf.write(output_path, np.array(audio, dtype=np.float32), samplerate=24000)
+    print(f"✅ Speech synthesis complete.")
+    print(f"> Final audio saved to: {output_path}")
+    
+    return output_path
 
 def match_audio_duration(original_path, translated_path, output_path="temp_audio_synced.wav"):
     """