Contents 1. Introduction
Feature extraction of speech is one of the most important issues in the field of speech recognition and representative of the speech. In the design of any speech recognition system, the best parametric representation of acoustic signals would be extracted and selected are important tasks. It significantly affects the recognition performance, a set of mel-frequency cepstrum coefficients (MFCC) provide a compact representation that are the results of a cosine transform of the real logarithm of the short-term energy spectrum expressed on a mel-frequency scale [1]. In recent studies of speech recognition system, the MFCC parameters perform better than others in the recognition accuracy [2], [3]. The Mel-Frequency Cepstral Coefficients (MFCC) is the most widely used features in speech recognition field. MFCC exploit the property of human auditory system (HAS), may acquire approximately perfect speech signal feature parameters in almost all possible voice transform spaces. Recent years, Chirp Z-Transform(CZT) gets more and more applicability and effectuality in signal processing field due to its highly selectivity of frequency range and resolution. In particular, the cepstrum spectral resolution can be improved by applying Chirp Z-Transform. In this paper, we performed the experiments; the results show the correctness and effectiveness of the MFCC and the CZT-based cepstrum in speech recognition for Mandarin digits recognition.
