Abstract: The automobile electric seat memory box need to determine the status of the work by the sound interval of the buzzer before they leave the factory,ear recognition often misjudge the missing problem due to the interference of external environment.In order to solve this problem,the computer is used to collect the sound of the buzzer,and voice is detected by using dynamic variable window length with double threshold endpoint.The short-time energy with MFCC coefficient as the sound characteristic parameters are extracted into the improved DTW algorithm and template matching.Experimental results show that the system can identify the target sound and have good robustness under the condition of noise.

Key words: car seat memory box, sound identification, automatic detection, improved endpoint detection, improved dynamic time warping