一种基于离散时延的鲁棒声源三维定位方法

Robust speech source 3D localization method based on discrete time delay

为了减少相位变换加权的可控响应功率(SRP-PHAT)声源定位算法的计算量,提出一种基于离散时延的改进算法.该方法首先利用FFT将麦克风阵列的每一帧接受信号变换到频域,然后在频域补零至16倍帧长,再运用IFFT将所有麦克风对的广义互相关函数在搜索之前计算好,从而可大幅度减少计算量.频域补零提高了广义互相关函数的采样率,因而由时延离散带来的定位误差很小.仿真结果表明,无论在远场还是近场条件下,该算法均能将计算量降低一个数量级而保持原算法的鲁棒性.

To reduce the computation load of the steered response power-phase transform（SRP-PHAT） which is a robust speech source localization algorithm,an improved SRP-PHAT algorithm based on discrete time delay is presented in this paper.In this method,a frame of signal from microphone arrays is transformed into frequency domain by FFT（fast Fourier transform）,then the sample points increase by 16 times by padding zeros in frequency domain.As a result,a generalized cross-correlation（GCC） of higher sampling rate can be achieved by taking IFFT（inverse fast Fourier transform）.All the GCCs can be calculated before searching;the computation load will be significantly reduced.Moreover,the localization errors introduced by discrete time delay are small enough to ignore because of the high sampling rate of GCC.Simulation results show that the method can save computation load by one order of magnitude,while still remaining robust in both far-field and near-field.