Custom Voice Signal Endpoint Detection Implementation

The method employed in this paper utilizes a custom-developed voice signal endpoint detection program that implements calculations for both short-term energy and short-term zero-crossing rate. The program determines voice signal endpoints by computing these two key acoustic features, enabling effective detection of voice signals. Short-term energy refers to the total signal energy within a specific time frame, typically calculated using frame-based processing where each audio frame's squared amplitude values are summed. Short-term zero-crossing rate measures how frequently the signal crosses the zero axis within a given timeframe, implemented by counting sign changes between consecutive samples in each processing window. By integrating these dual features through a weighted decision algorithm, the program achieves more precise identification of voice signal start and end points, thereby enhancing the accuracy and stability of voice endpoint detection. The implementation typically involves frame segmentation, feature extraction using sliding windows, and threshold-based classification logic to distinguish between speech and non-speech segments.