The Research And Implementation Of Automobile Seat Memory Box Detection System Based On Voice Recognition

Memory box for automobile seat is the core component of automatic adjustable seat and needs to be performed qualification testing before sell.At present,the workers on the production line distinguish the product qualification according to the memory box buzzer sounds internal.If the product is qualified,the buzzer will turn three sounds;if not issued three sound clear ringing,it failed the test results.Due to the interference of external environment,ear recognition often misjudge the results.In this paper,the computer is used to design a set of system which can detect the validity of the memory box by using the technology of sound recognition.First of all,the detection is performed in order to solve the problem of dividing the target signal and the background noise by using the difference of the average amplitude and zero crossing rate for the noise and the beep under the condition of SNR of 15~30dB.The length of the window is automatically variable,which improves the detection accuracy and speed effectively.The detection is performed by using the difference of spectral entropy between noise and the beep under the condition of SNR of 0~15dB.The target signal and noise can be accurately distinguished in low SNR environment.Secondly,the distribution of the traditional Mel filter banks is improved according to the characteristics of the buzzer and the new Mid-Mel filter bank is used to improve the resolution of the target signal.The MFCC coefficient and its difference spectrum that mixed with the TEO energy are extracted as the feature vector for target signal and the large scale vector is selected by the Fisher criterion.Three kinds of mixed characteristic parameters were formed after optimization.It is of great significance to grasp the characteristics of the beep accurately for improving the recognition rate in feature extraction.Finally,the DTW algorithm is choosed for target sound in this paper and the search path of DTW is improved to reduce the recognition time.In order to solve the problem that the algorithm is sensitive to the accuracy of endpoint detection,the algorithm is improved by relaxing endpoint.The results show that the improved algorithm has better robustness under noise conditions.The improved algorithm performance is compared with the results of artificial hearing recognition and it is found that the machine recognition is superior in stability and accuracy.