http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
고립단어의 선형 예측 계수들 상호간의 상관도를 이용한 화자인식에 관한 연구
오예환,김중규 成均館大學校 科學技術硏究所 1995 論文集 Vol.46 No.2
We propose a statistical algorithm which is applicable for speaker recognition and identification. The decision logic is formulated by computing the correlation coefficient between two speakers based on the LPC Log Spectrum and the LPC based Cepstral Spectrum of the speech signals. We tested the algorithm with Korean word "Sung Kyun Kwan" spoken by 11 male speakers, which is sampled at 16KHz rate and quantized to 16 bits. The test result showed very high recognition rates(94.5% for the case of LPC Log Spectrum and 98.2% for LPC based Cepstral Spectrum), whereas the conventional DTW algorithm based on the pitch contours of the speech signal had a relatively low recognition rate(89%). In addition to the recognition rate, the correlation method outperformed the DTW method in processing time as well, i.e. it requires much less computation load than DTW. Thus the proposed algorithm seems to have a strong potential to be applied to the areas where very high recognition rates as well as real time processing are required such as a security control system.