| Summary: | 碩士 === 國立中興大學 === 統計學研究所 === 99 === This paper is to investigate the Chinese consonant recognition given that the vowels are correct. We use the feature of linear predictive cepstrum coefficient and the method of K-nearest neighbor for the speech recognition of 337 isolated mandarin words. Four experimental factors in this paper we considered. That is , the number of sampling points in each frame,“the dimension of cepstrum parameters, the number of shifting frames, and the shifting duration of the consonant. We try to find the optimal parameters for the recognition, and the best rate of recognition is 86.75% in this paper.
Keywords: Linear predictive coding、K-nearest neighbor
|
|---|
