基于次梯度投影的数字助听器自适应声源定位方法

Acoustic source localization based on adaptive subgradient projection in digital hearing aids

该方法在特征值分解算法的基础之上,利用次梯度投影方法自适应估计声源到麦克风的脉冲响应系数,进而估计出各麦克风之间时延,并利用几何方法定位声源在3D空间的位置.与传统的基于广义互相关的时延估计算法相比,提出的算法在房间反射与共振的情况下定位精度更高;与基于NLMS算法的自适应特征值分解时延估计算法相比,提出的算法收敛速度更快,并且在强噪声的情况下鲁棒性更强.基于眼镜数字助听器声源定位系统的实验与仿真研究了麦克风阵不同的几何尺寸对算法性能和定位精度的影响,证明了在不同信噪比情况下该算法都能有效定位声源的3D空间位置.

Based on the eigenvalue decomposition （EVD） algorithm, the proposed method estimates the impulse response coefficients between speech source and microphones by means of adaptive subgradient projection algorithm, then acquires the time delays of microphone pairs, and calculates the source position in 3D space by geometric method subsequently. Compared with the traditional time-delay estimation algorithms based on generalized cross-correlation （GCC）, the proposed method achieves more accurate results when reverberation exists. Compared with the adaptive normalized least mean squares-EVD algorithm, the proposed method converges faster and is more robust under strong noises. Experiments and simulations based on glasses hearing aid show the influences on the localization performance for different microphone array sizes, and demonstrate the validity of the proposed method using signals with different signal-to-noise ratios （SNRs）.