| Crying is an important way for the infants to communicate with the world,the infants' crying contains a lot of physical and psychology information,it reflects that they need some kind of help,the infants' crying is a kind of alarm to the outside world.Monitoring the infants' crying has great significance.Firstly,we designed the hardware and software for the embedded system,we used the RT5350 chip as the processor which is based on the MIPS architecture,we ported Linux kernel on the chip and programmed the C language programs of the algorithm.The embedded system can detect the infants' crying on real time.Secondly,we discussed the pretreatment and characteristic parameters extracting of infants' crying.The 6-order,12-order,and 24-order MFCC were extracted in the experiment.We used the dynamic time warping(DTW)algorithm as recognition algorithm and the algorithm was optimized.We established an infants' crying library which contains 130 infants' crying samples.It was divided into three groups,the first group which contains 60 samples was used to obtain the reference templates,and the second group which contains 40 samples was used to test the recognition rates of the algorithm,the third group which contains 30 samples was used to obtain the threshold.We used percussion,door slamming,and human voices to do the comparative tests.We conducted experiments by extracting 6-order,12-order,24-order MFCC and found that the 12-order MFCC worked best.We conducted the experiment by using the dynamic time warping algorithm and the improved dynamic time warping algorithm respectively.It proved that the improved dynamic time warping algorithm can not only improve the recognition rates of the infants' crying,but also more efficient.By using 12-order MFCC as characteristic parameters and using the improved dynamic time warping algorithm as recognition algorithm,the recognition rates of the algorithm was 90% for the testing set,and the algorithm is robust for the noise. |
PDF Full Text Request