電子情報通信学会ソサイエティ大会講演要旨
A-15-10
Robust Speech Recognition using MFCC with Triangular Mel Filtered Time Varying LPC
○George Mufungulwa・Hiroshi Tsutsui・Asheralieva Alia・Yoshikazu Miyanaga(Hokkaido Univ.)
This paper proposes a new ASR with MFCC of TVLPC. In the proposed approach, triangular mel filtered time-varying LPC features are appended to FFT based MFCC features. Evaluation results of similar pronunciation phrases experiments on 15 types of noise demonstrate that the proposed approach achieves better recognition accuracy at 62.67\% with CMS/DRA at 10\,dB and 71.78\% with RSA at 20\,dB respectively. Although the approach can be used for any phases, it is only useful under noisy condition.