基于相关峰细化时延估计的舱内服务机器人发话人定位研究

发话人定位是舱内服务机器人有效区分航天员与环境并获取与航天员相对位置关系的重要手段。针对时延估计的机器人声定位算法精度受采样频率和噪声限制的问题,提出了一种基于相关峰精确插值的空间六元阵列发话人定位方法。该方法基于机器人球形结构设计,利用信号预处理和二次相关降低噪声干扰,通过线性调频Z变换(MCZT)取代快速傅里叶变换(FFT)计算细化频谱,突破已有时域采样率的限制,能有效弱化FFT带来的栅栏效应,提高相关函数分辨率、时延估计精度及发话人定位精度。实验结果表明:基于相关峰精确插值算法的发话人定位方法,其时延估计性能有明显提升,能较好地对声源目标进行定位,且算法精度优于基于广义互相关的发话人定位方法,能满足舱内服务机器人的定位需求。

Study on Speaker Localization for Assistant Robots in Spacecraft Based on Fine Interpolation of Correlation Peak

Speaker localization is an important means for assistant robot in spacecraft to effectively distinguish the astronauts from the environment and acquire the relative position relationship with the astronauts. To solve the problem that the accuracy of robot sound localization algorithm based on time delay estimation (TDE) is limited by sampling frequency and noise, a speaker localization algorithm based on six-element cone-shaped array via fine interpolation of correlation peak (FICP) is proposed in this paper. The improved method based on the spherical structure uses signal pre-processing and secondary correlation to reduce the noise, and uses the modified chirp-Z transform (MCZT) rather than the fast Fourier transform (FFT) to calculate the refined spectrum for breaking through the limitation of the existing time domain sampling rate. The improved method can effectively weaken the fence effect caused by FFT. It can raise the resolution of correlation function and improve the accuracy of TDE and speaker localization. Experimental results show that the speaker localization method based on FICP algorithm can locate the sound source target efficiently, and the accuracy of the improved method is better than that of the method based on generalized cross-correlation. Thus, the method can satisfy the localization requirement of the assistant robot.