Evaluation and design of wavelet packet cepstral coefficient (WPCC) for a noisy Indonesian vowels signal
Syahroni Hidayat, Abdurahim Abdurahim, Muhammad Tajuddin · Journal of Physics Conference Series · 2019
Wavelet feature,Wavelet Packet Cepstral Coefficient (WPCC),is widely used both in speaker and speech recognition systems. WPCC is designed to replace the role of MFCC. Unfortunately, WPCC still needs to be developed primarily in the process of extracting a noisy signal.Since the possibility of noise mapped into nodes as a result of wavelet decomposition and being a feature coefficient is very high. Therefore, this study aims to analyze the raw design of WPCC filters which extracted from a noisy signal.Mean Best Basis (MBB)algorithm with wavelet function db44 and db45 is applied to obtain it. The results show, based on the average SNR value, all data used in this study categorized as noisy signals. The implementation of MBB with db44 and db45 wavelet function able to generate the raw design of WPCC filter. Only two types of raw design created. Both have a different number of nodes, each with 59 nodes and 37 nodes. The noise does not influence this difference,but the effect of the entropy usage, the spectral properties of Indonesian vowel signals, and the wavelet function applied.