Comparison of Various Windowing Methods on Filter Bank for Speech Recognition

Mel-Frequency Cepstral Coefficients (MFCC) are one of the techniques used for extracting audio features. In the MFCC feature extraction process, there is a step called Mel-Frequency Wrapping, which is based on a band-pass filter. Typically, this band-pass filter uses a triangular filter. This paper proposes to compare the triangular filter with several windowing methods in the Mel filter bank. The windowing methods chosen for comparison are the Hamming Window, Hanning Window, and Blackman Window. The data used in this study consists of audio recordings of the pronunciation of the numbers 0 to 9 in Bahasa Indonesia. The results show that the Blackman window achieves the highest accuracy at 96.43% with a reduction in standard deviation and average MCC at 0.96.