论文部分内容阅读
为了提高语音端点检测的准确性,增强端点检测算法在噪声环境下的鲁棒性,提出两种新的端点检测参数。其中,基于临界频带的谱熵参数综合考虑了人耳对语音的感知特性以及语音信号和噪声信号的频域分布差异,差值频域能量参数考虑了语音帧和无声帧在频域上的能量差异。结合两种参数的优点,构成一种鲁棒的端点检测参数,同时,为了避免因阀值判决的单一性而产生误判,在端点检测过程中加入了基于特征分布统计的过渡段判决。试验结果表明,本研究提出的语音端点检测算法对语音帧和无声帧具有较好的区分性,在不同噪声且信噪比较低情况下,端点检测准确率相比传统抗噪端点检测算法均有所提升,特别是在非平稳噪声下,准确率提升超过5%。
In order to improve the accuracy of voice endpoint detection and enhance the robustness of endpoint detection algorithm in noisy environments, two new endpoint detection parameters are proposed. The spectral entropy parameters based on the critical frequency band take into account the perceived characteristics of the human ear as well as the frequency domain differences of the speech signal and the noise signal. The difference frequency domain energy parameter considers the energy in the frequency domain of the speech frame and the silence frame difference. Combining the advantages of the two parameters, a robust endpoint detection parameter is constructed. In order to avoid misjudgment caused by the singleness of the threshold decision, a transition decision based on the feature distribution statistics is added during the endpoint detection. The experimental results show that the speech endpoint detection algorithm proposed in this study has a good distinction between speech frames and silence frames. Compared with traditional anti-noise endpoint detection algorithms, the accuracy of endpoint detection under different noise and low signal-to-noise ratio The improvement has been made, especially in non-stationary noise, with an accuracy rate of more than 5%.