基于子带无迹粒子滤波的语音增强算法

Subband unscented particle filter for speech enhancement

结合多采样率系统理论中的子带分解技术与贝叶斯估计理论中的无迹粒子滤波技术,提出了一种基于子带无迹粒子滤波的语音增强方法。该方法首先将语音信号分解成子带信号,建立各子带信号的低阶时变自回归模型;然后利用无迹粒子滤波估计模型参数,对子带信号进行滤波处理;最后根据滤波后的子带信号重构语音信号,实现语音增强。仿真结果表明,该方法能明显改善语音信号的信噪比和质量,且易于实现。

A novel method was proposed for speech enhancement based on unscented particle filter（UPF） and subband decomposition.First of all,the noisy speech is decomposed into subband speech signals.And then these signals are modeled as low-order time-varying autoregressive（TVAR） processes.Secondly,unscented particle filters are applied to estimate the parameters and process the subband speech signals.Finally,the enhanced fullband speech signals are reconstructed from the enhanced subband speech signals.Simulation results show that the proposed method can improve obviously signal-to-noise ratio and the quality of speech,and it is easily realized on real time.