基于自适应参数估计的鲁棒语音增强算法

Robust speech enhancement with adaptive parameter estimation

基于鲁棒H∞滤波器理论和共轭梯度自适应参数估计方法提出了一种对复杂噪声有抑制效果的语音增强算法。应用这种方法自适应地从带噪信号中提取语音参数时不必预先知道噪声源的统计特性,只要求噪声信号能量有限。因为它基于H∞滤波器,所以可保证由外界干扰和附加噪声引起的性能指标恶化达到最小。仿真结果表明:该语音增强算法具有计算速度快、鲁棒性好、语音增强效果明显、易于实现、可抑制复杂背景噪声等特点。

A new speech enhancement approach based on H∞ filtering and adaptive conjugate gradient parameter estimation is proposed.In this approach no a priori information of noise statistics is required when conducting adaptive speech parameter estimation from noised speech.one assumption needed on the noise is that noise signal has finite energy.on the basis of H∞ filter, minimization of the performance index deterioration of this algorithm caused by additive noise is guaranteed.Simulation results show that this speech enhancement algorithm is fast,robust,effective,easy to implement with single speech input channel,and can restrain complicated background noise.