Energy Consumption Analysis and Optimization of Speech Algorithms for Intelligent Terminals

Aim: ing at the rapid popularization of voice assistants in national defense, aging healthcare and public management, this paper systematically analyzes the arithmetic ratio and energy consumption bottleneck of the algorithms at each stage, and proposes lightweight neural network pruning and compression strategies. Without sacrificing the recognition accuracy, the overall reasoning energy consumption is reduced by 42%, and the integrity of the algorithm is verified by the hardware trusted execution environment, which is in line with the national strategic demand of "green AI" and safe and controllable.