Estimation of sound pressure levels of direct sound and reflected sounds based on single-channel signal separation

摘要

Recent studies have shown that the reflected sounds from nearby buildings affect the speech intelligibility of public address (PA) systems. Therefore, the PA system volumes and the loudspeaker directions should be adjusted to compensate for the effect of reflected sounds which can change over time due to building construction. This study aims at estimating the sound pressure levels (SPLs) of direct sound and reflected sounds that can be taken as a consideration before adjusting the PA systems. This paper proposes an approach for separating direct sound and reflected sounds that employs time-frequency information of a single-channel signal. This method consists of three steps: (1) the estimation of time difference of arrivals (TDOAs) between direct sound and reflected sounds, (2) the estimation of reflected-to-direct gain ratios (RDRs), and (3) the single-channel signal separation based on (1) and (2) in frequency domain. Numerical tests were conducted to evaluate the performance of the proposed method. The results show that the proposed method is able to separate direct sound and up to five reflected sounds as well as estimate its SPLs.