Blind estimation of reverberation parameters for non-diffuse rooms

Reverberation time is an important objective parameter for the acoustics of enclosed spaces. Blind estimation of the reverberation time from naturally occurring sound sources alleviates the logistical constraints of traditional measurement methods, and hence enables non-invasive, in-situ measurements in occupied spaces. A maximum likelihood method using a simple exponential decay model was recently developed. It demonstrated the potential of this approach, but the decay model hinders its accuracy and usefulness in the presence of more complex decays. This paper proposes a new maximum likelihood approach that allows for the multiple decay rates found in real spaces to be more correctly modelled, and thus improves the accuracy of the estimations. Envelopes of received sound signals are segmented and suitable decay phases automatically selected. A maximum likelihood estimation is performed using a multi-decay model whose function adapts to yield the most likely reverberation parameters and decay curve estimates. Simulation, validation and real room tests confirm the improved accuracy; the estimation errors are typically below perceptual difference limens. Thus a new method for blind estimation of reverberation parameters from naturally occurring sound sources, including speech and orchestra music, is achieved