Speech-Based Five Features Extraction and Hardware Trade-Off Design for Mental Fatigue Monitoring

After combing the known advantages and disadvantages of the normal methods on human mental fatigue test, we measure speech signals in condition of fatigue-free and fatigue, then attract key features, and then analyze their validities. More interesting works address corresponding feature-modules designed with EDA tools. The methods are outlined as: 4~5 seconds data are sampled by PC embedded MATLAB software platform, the mean-value ratio of pre- and post-features is defined into evaluation threshold. And with the help of EDA tools named of System Generator and Xilinx ISE 11, our feature-modules design optimization are marched onto complexities' report. The results show that five key features' validity-sequence can list as (1) short-time energy (mean-bias 298%), (2) short-time mean amplitude (41.9%), (3) short-time cross-zero ratio (15.9%), (4) short-time self-correlation (12%), and (5) pitch (10%), just about same as their modules' structure-sophistications. Compared with other advanced tricks on mental fatigue tests, our works may accelerate low power consumption application combined with trade-off design thinking for future study of cerebral health monitoring.