Mixed Signal Approach Using C++ and DSP Hardware for Low Latency and Secure Speech Systems

Abstract

Real-time  speech  systems  need  to  provide  quick  responses  while  remaining  resistant  to security  risks.  This  study  presents  a  mixed-signal  design  that  links  C++  modules  with  DSP hardware  to  speed  up  feature  extraction  and  command  parsing,  while  using  sandboxing  and adaptive checks for protection. Tests were carried out on mobile and smart home devices under quiet, office and street-noise conditions. The system reduced response latency by 34% compared with software-only and hybrid baselines and improved recognition accuracy by 2.5–4.6 percentage points. Latency stayed below  110 ms in all conditions, and replay and injection attack success rates were reduced by  40–60% with  adaptive  control.  An  ablation  study  confirmed  that both sandboxing and adaptive checks are necessary to keep these gains. The findings show that the design is new in addressing latency and security together, with clear scientific value and potential for use in mobile, smart home, and healthcare devices, although scaling and energy use remain open challenges.