An adaptive IMM particle filter algorithm based on multi-feature for sound source tracking in reverberant and noisy environments

To deal with the problems of low accuracy and weak robustness of sound source localization in reverberant and noisy environments, an adaptive IMM particle filter algorithm based on the multi-feature is proposed. In this algorithm, the mechanisms of time delay selection and beam output energy optimization are established, by using the space-time correlation and iterative filtering, where the multi-features of the signals received by the microphones are exploited as the observation information. Subsequently, the reasonable sound source position information is obtained from the likelihood function, which is constructed on the basis of both. Meanwhile, considering the randomness of speaker motion, an adaptive IMM algorithm is given. By generating online particle set and interacting the models with different process variances, the speaker’s different motion modes are fitted, which improves the robustness of the speaker tracking system. The simulation and experimental results show that the complementarity of the location information based on the multi-feature is employed in the proposed algorithm, and the influence of the uncertainty of the observation error on sound source position estimation is reduced. Simultaneously, the robustness of random moving sound source tracking system is enhanced and the positioning accuracy of the system is improved. © 2023 South China University of Technology. All rights reserved.