Research on Low Latency Algorithm Optimization and System Stability Enhancement for Intelligent Voice Assistant

With the widespread integration of voice interaction systems in digital government services, emergency response terminals, and embedded smart devices, ensuring real-time responsiveness and system robustness has become a critical requirement. However, under highly concurrent interaction conditions, commercial voice assistants often suffer from excessive response latency and unstable memory behavior, particularly in constrained embedded environments. This study addresses these challenges by proposing a dual-path optimization framework that combines fine-grained memory lifecycle management based on Objective-C and a refined multi-threaded asynchronous scheduling mechanism. Built on an industrial-grade prototype, the optimized system introduces a memory tagging and automatic deallocation strategy along with a weighted task priority mapping algorithm, significantly enhancing response efficiency and stability. Experimental evaluations under 5000 concurrent voice requests demonstrate a reduction in average response latency from 410.2 ms to 284.7 ms and a crash rate decrease of $\mathbf{8 2. 3 \%}$. Compared with conventional strategies such as FastPath caching or GPU-only acceleration, the proposed approach achieves better consistency in scheduling, reduced resource contention, and improved resilience under failure conditions. These results support the scalable deployment of intelligent voice assistants in mission-critical and latency-sensitive applications.