声源定位通常是利用传声器阵列的到达时间差实现,时延估计算法的精度和鲁棒性对于定位结果非常重要。对时延估计中最常用的广义互相关法进行了详细研究,取不同的加权函数进行仿真,比较分析了仿真结果,并引入语音增强中的谱减法作为预处理环节,对传统的广义互相关法进行了改进。结果表明,新方法能满足噪声环境下的要求。
The acoustic source localization is usually accomplished by time delay of arrival of microphone arrays, and the precision and robust of time-delay algorithms is very important. Generalized cross correlation often used in time-delay estimation is studied in detail and simulated with different weight functions, and the results are compared. Then the spectral subtraction method used in speech enhancement is introduced to be pretreatment, and the generalized cross correlation is improved. The simulation is accomplished to prove the effectivity of modified algorithm. The result shows that it can satisfy the demand of noise environment.