Voice Signal Endpoint Detection and Pitch Trajectory Detection

In this paper, we address speech signal processing tasks including endpoint detection and pitch trajectory detection for input voice signals. Endpoint detection can be implemented using various methods such as autocorrelation function (ACF), average magnitude difference function (AMDF), and combined ACF/AMDF approaches. The ACF method computes the correlation between a signal and its time-shifted version to identify periodic patterns and voice activity regions. AMDF calculates the average magnitude difference between the original signal and its delayed version, effectively detecting signal energy variations at utterance boundaries. The combined ACF/AMDF approach utilizes the strengths of both methods for improved robustness. For pitch trajectory detection, these functions help track fundamental frequency variations over time by analyzing periodicity in voiced segments. Implementation typically involves frame-based processing with overlapping windows, threshold-based decision making, and peak detection algorithms to identify pitch periods and speech endpoints. These methods enable comprehensive analysis and processing of speech signals for applications like speech recognition and voice analysis systems.