混响背景下GTM回波检测算法研究

为改善混响背景下传统匹配滤波算法效果不佳问题,在分析其非平稳性、有色性和非高斯性的基础上,提出了混合高斯时变自回归模型(Gaussian mixture Tvar Model,GTM),推导了模型公式及其参数求解方法,形成了GTM回波检测算法。为对混响特性及滤波效果进行定量描述进而验证算法性能,给出了一种定量衡量混响非平稳性、有色性、非高斯特性的滤波效果评价方法。通过实测混响分析表明,GTM模型能够较好地拟合实测混响的概率密度曲线和功率谱密度曲线,实现了混响背景下回波的有效检测并改善混响特性。

Research on GTM echo detection algorithm under reverberation background

Traditional matched filtering algorithm does not work well under the reverberation background, On the basis of analyzing non-stationarity, colority and non-Gaussian of reverberation, this paper presents a Gaussian mixture time-varying autoregressive model (GTM for short), derives the model formula and parameter solving method, and then establishes the GTM echo detection algorithm. To describe reverberation features and filtering effect quantitatively, a reverberation feature evaluation method is presented. According to the analysis of measured reverberation, the GTM model can better fit the measured probability density and power spectral density curves, and effective echo detection can be done by this algorithm under the reverberation background, meanwhile the reverberation features are also improved.