融合Burg谱估计与信号变化率测度的语音端点检测

发布时间：2019-05-30 05:03

【摘要】：针对现有基于特征的语音端点检测方法在低信噪比及非平稳噪声下检测性能较低的问题,提出了一种融合Burg谱估计与长时段信号变化率测度(LTSV)的语音端点检测方法.该方法采用表征较长时段语音变化率的LTSV参数,较准确地反映了语音的非平稳程度.与传统基于特征的语音端点检测方法相比,该方法在低信噪比及非平稳噪声情况下的检测性能有了较大提高.并融合Burg谱估计,与传统Welch谱估计方法相比,提高了LTSV参数的区分度,从而进一步提高了检测的准确率.仿真结果表明:采用融合Burg谱估计与LTSV的语音端点检测方法在低信噪比(-10dB)及非平稳噪声情况下,与传统基于特征的语音端点检测方法相比,检测准确率普遍提高了约6%以上,说明该方法在低信噪比及非平稳噪声环境下鲁棒性更好.
[Abstract]:In order to solve the problem that the existing feature-based speech endpoint detection methods have low detection performance under low signal-to-noise ratio (SNR) and non-stationary noise, a speech endpoint detection method combining Burg spectrum estimation and long-term signal change rate measure (LTSV) is proposed. In this method, the LTSV parameters are used to characterize the speech change rate for a long time, which accurately reflects the non-stationary degree of speech. Compared with the traditional feature-based speech endpoint detection method, the detection performance of this method is greatly improved under the condition of low signal-to-noise ratio (SNR) and non-stationary noise. Compared with the traditional Welch spectrum estimation method, the discrimination of LTSV parameters is improved by fusion Burg spectrum estimation, which further improves the accuracy of detection. The simulation results show that the speech endpoint detection method based on fusion Burg spectrum estimation and LTSV is compared with the traditional feature-based speech endpoint detection method in the case of low signal-to-noise ratio (- 10dB) and non-stationary noise. The detection accuracy is generally improved by more than 6%, which indicates that the method is more robust in low SNR and non-stationary noise environments.
【作者单位】：西北工业大学电子信息学院;
【基金】：陕西省自然科学基金资助项目(2011JQ8038)
【分类号】：TN912.3

【参考文献】