A Hybrid Method for Blind Estimation of Frequency Dependent Reverberation Time Using Speech Signals

Reverberation time is an important room acoustical parameter that can be used to identify the acoustic environment, predict speech intelligibility and model the late reverberation for binaural rendering, etc. Several blind estimation algorithms of reverberation time have been proposed by analyzing recorded speech signals. Unfortunately, the estimation accuracy for the frequency dependent reverberation time is lower than for the full-band reverberation time due to the lower signal energy in sub-band filters. This study presents a novel approach for the blind estimation of reverberation time in the full frequency range. The maximum likelihood method is applied for the estimation of the reverberation time from low- to mid-frequencies, and the reverberation time from mid- to high-frequencies is predicted by our proposed model based on the analysis of the reverberation time calculated from room impulse responses in different rooms. The proposed method is validated by two experiments and shows a good performance.