Speech Recognition for Digits 0-9 Based on MATLAB

In this study, we implemented speech recognition for passwords composed of digits 0-9 by calculating the cross-correlation function between signals and their corresponding variances. To achieve higher accuracy, signal preprocessing is essential. This includes applying FIR filters for pre-emphasis and extracting useful signals through endpoint detection based on short-term average energy and zero-crossing rate methods. The preprocessing steps enhance signal quality and reliability, thereby improving speech recognition accuracy. In MATLAB code, this typically involves designing FIR filters using fir1() function with appropriate cutoff frequencies, and implementing frame-based processing where energy and zero-crossing rates are calculated for each windowed segment to detect speech endpoints.

The cross-correlation function, which measures similarity between two signals, helps distinguish test signals from template signals. By employing maximum variance as a discrimination threshold, we can effectively identify digital passwords within speech signals. The implementation uses xcorr() function to compute cross-correlation and var() function for variance calculation, followed by threshold comparison logic. Overall, this research adopts a series of effective methods aimed at enhancing speech recognition accuracy and achieving reliable digital password identification through systematic signal processing and pattern matching techniques.