论文提出一种适用于实际噪声环境的语音端点检测算法,并将其用于语音识别系统中。首先通过跟踪平稳噪声能量,检测能量变化非平稳的信号;然后在这段信号中根据浊音特性搜索并跟踪谐波,确定浊音的位置:最后结合能量和浊音信息检测语音的精确起止点。实验证明,该算法能在多种环境噪声中保持稳定可靠的性能。
This paper suggests a new method of endpoint detection for real noise environments.First,the noise energy, is tracked to detect the unstable signal.Then the harmonics of voice are detected and tracked to get the position of voice. At last,the refined speech endpoint is detected based on energy and voicing features.The new method operates reliably in most real noise environment.