论文部分内容阅读
语音端点检测是语音处理中重要的领域之一。常规谱熵语音端点检测算法是通过检测语音的功率谱的平坦程度,从而达到语音端点检测的目的。但是该方法在平稳噪声环境下较好,在无噪声和非平稳噪声环境下效果较差。作者在分析了无噪声环境下常规谱熵端点检测算法效果差的原因的基础上,结合了语音的短时能量算法,对常规谱熵算法进行了改进,形成了一个新的特征参数——谱熵能量积。仿真结果显示,该方法相对于常规谱熵算法,在无噪声的环境下检测精度有了很大的提高,在非平稳噪声环境下也有了一定的提高,鲁棒性得到增强。
Voice endpoint detection is one of the most important areas in voice processing. Conventional spectral entropy voice endpoint detection algorithm is through the detection of the flatness of the power spectrum of speech, so as to achieve the purpose of voice endpoint detection. However, this method is better in stationary noise environment and less effective in no-noise and non-stationary noise environment. Based on the analysis of the poor performance of conventional spectral entropy endpoint detection algorithm in no noise environment, this paper combines the short-time speech energy algorithm to improve the conventional spectral entropy algorithm and forms a new characteristic parameter-spectrum Entropy energy product. The simulation results show that compared with the conventional spectral entropy algorithm, the proposed method can greatly improve the detection accuracy under no noise environment and improve the robustness under non-stationary noise environment.