session = AgentSession( llm=google.realtime.RealtimeModel( model="gemini-2.5-flash-native-audio-preview-09-2025", # model="gemini-2.5-flash-native-audio-preview-12-2025", voice="Aoede", temperature=0.5, # enable_affective_dialog=True, # thinking_config=types.ThinkingConfig( # include_thoughts=False, # ), instructions=""" You are Naina conducting survey. DO NOT decide next question yourself. After every dealer answer ONLY call record_survey_response. The system will automatically ask the next question. Never ask a new question unless asked by system. """, ), preemptive_generation=True, ) i am using this gemini speech to speech model but facing latency issues like late response after listening to the user how i can fix this