Integrated Memory Control and Thread Scheduling for Real-Time Voice Interaction Systems

Abstract Intelligent voice assistants need to keep response delay low and service stable while running on devices with limited hardware resources. This study used a dual-path method that joins fine memory control with multi-thread asynchronous scheduling. A total of 118 runs were carried out on smartphones, embedded boards, and wearables under both controlled and office settings. The results showed that median response time fell by 27% and 99th-percentile delay fell by 34%. System throughput rose by 22%, and stability improved by about 30%. Accuracy stayed steady, with word error rate changes within 0.2 and F1 score changes within 0.3. Processing cost was reduced by about 42% compared with standard models. These outcomes show that delay and stability can be improved together without loss in accuracy. The method can support faster and steadier voice assistants, though more devices, longer tests, and far-field or multi-language data should be included in future work.