Comparative Performance Analysis for Speech Digit Recognition based on MFCC and Vector Quantization

Abstract The main goal of this research work is to experimentally verify the importance of spoken Speech digit signal in person authentication in controlling applications. The motivation is based on the earlier work of demonstrating the feasibility of using spoken speech digit utterance signal for person security and controlling applications. This paper work also discusses the. Comparative analysis of the cepstral analysis with the mel frequency cepstral coefficient (MFCC) by using vector quantization feature matching technique. All digits speech digit from zero utterance to nine digit utterance data has been collected for 15 subjects in three different sessions. For the thus collected spoken speech digit data, the feature extraction techniques such as cepstral and MFCC were applied to extract the Cepstral and MFCC features. In the next stage of work vector quantization was used for feature matching for both Cepstral and MFCC features and performance were recorded for two different session data. By comparing the performance of Cepstral plus VQ with the MFCC plus VQ, we can conclude that feature extraction technique MFCC gives the better performance than cepstral feature for spoken digit utterance data.