Influence of LPC Order on the Voice Comparison

• Main Author: 蔡文忠
• Other Authors: 徐健民
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/quzfsp

碩士 === 中央警察大學 === 鑑識科學研究所 === 90 === As the progressing of technology, the communication among people is more convenient, and the speech recognition has been the important clue to investigate crimes. In the thesis, the speech signal of the same person and the different persons were compared with the normal factor of speech recognition system and two characteristic factors named linear predictive code and cepstral coefficient. By comparing with the statistic means and standard deviations, the resolution due to different sampling of speech digitalizing was calculated and then decided the optimal orders of linear predictive code and cepstral coefficient which led to the operational criterion of voice comparison The sampling of speech digitalizing was determined by the highest frequency of the record system. The excessive sampling would not help to raise the resolution but spend more times in operation. The ideal resolution was obtained while the speech sample with 8000Hz sampling was filtrate by 3500Hz low pass filtering and the orders of cepstrum and LPC were set at 18 and 12 respectively. The resolution will be lowered if the speech sample is not compared on the moment. The disadvantage of resolution caused by time lapse can be recovered by comparing the whole sentences composed of three to five words.