Frequency Domain Analysis of MFCC Feature Extraction in Children’s Speech Recognition System

Abstract —The research on speech recognition systems currently focuses the analysis of robust systems. When signals are combined with noise, system becomes distracted, struggling to identify sounds. Therefore, development a continues be carried out. The principle is eliminate noise from and restore original information signals. In this paper, researchers conducted frequency domain one stage Mel Frequency Cepstral Coefficients (MFCC) process, Fast Fourier Transform (FFT), in children's system. FTT feature extraction process determined effect value characteristics utilized FFT output disruption. method was designed into three scenarios based employed points. differences between were number shared All points divided four, three, two parts first, second, third scenarios, respectively. This study data isolated TIDIGIT English digit corpus. As comparative data, added manually simulate real-world conditions. results showed that using particular portion following scenario MFCC affected performance, which relatively significant noisy data. 3 (C1) version generated highest accuracy, exceeded accuracy conventional method. average increased by 1% more than all tested types. Using various intensity values (SNR), testing indicates generates higher SNR values. It proves selection specific significantly affects speech.