Update app.py

app.py

@@ -80,38 +80,42 @@ def ask_openai_with_rag(query):
         traceback.print_exc()
         return "[Error: Failed to generate response.]"
 
-def synthesize_speech(text, lang="en"):
+def synthesize_speech(text):
     try:
-        tts_pipeline = tts_models["en"]
-        output = tts_pipeline(text)
-        audio = output["audio"]
-        # Ensure it's a numpy array (required by Gradio)
-        if isinstance(audio, np.ndarray):
-            return (audio, 22050)
-        else:
-            print("Unexpected audio type:", type(audio))
-            return None
+        # Always using English TTS
+        output = tts_models["en"](text)
+        audio = output["audio"]  # usually float32 numpy array
+
+        # Normalize to int16 PCM
+        max_val = np.max(np.abs(audio))
+        if max_val > 0:
+            audio = audio / max_val
+        audio_int16 = (audio * 32767).astype(np.int16)
+
+        return (22050, audio_int16)
+
     except Exception as e:
         print("TTS error:", e)
         return None
 
 
+
 def full_pipeline(audio):
     transcription = transcribe_audio(audio)
-    lang = detect_language(transcription)
 
     if "[Error" in transcription:
         return transcription, "Sorry, I couldn't understand that.", None
 
     reply = ask_openai_with_rag(transcription)
-    tts_audio = synthesize_speech(reply, lang)
+    tts_audio = synthesize_speech(reply)
 
     # Ensure we don’t pass invalid audio to Gradio
-    if not isinstance(tts_audio, tuple) or not isinstance(tts_audio[0], np.ndarray):
+    if not isinstance(tts_audio, tuple) or not isinstance(tts_audio[1], np.ndarray):
         return transcription, reply, None
 
     return transcription, reply, tts_audio
 
+
 # ---- GRADIO UI ----
 with gr.Blocks() as demo:
     gr.Markdown("# 🏠 Voice-Based Real Estate Assistant (Hindi + English)")