抄録
This paper aims at achieving robust direction-
of-arrival (DOA) estimation under noisy conditions. To improve
the noise robustness, a confidence measure is introduced to
estimate the accuracy of each DOA estimate and to adaptively
switch over its algorithm depending acoustic environments. DOA
estimation is carried out by particle filtering on spatial state
space with a target source model and an environmental noise
model. The target source model gives a system model in the
framework of state estimation, and the noise model tries to find
the dominant frequency of the target signal with noisy
observation to prepare a reliable likelihood. The proposed
confidence measure is defined as effective sample size in particle
filtering, and determines whether the noise model should be
updated time by time. In this paper, we discuss the suitability of
the effective sample size as a confidence measure for DOA
estimation, and show the effectiveness with the performance of
DOA estimation under a noisy condition.
