Using long-term coherence estimates for binaural de-reverberation

A novel binaural de-reverberation algorithm is presented which utilizes the properties of the interaural coherence (IC) inspired by the concepts introduced in Allen et al. [1]. Studying the long-term IC statistics revealed a close link between IC and source/receiver distance, room reverberation time and source directivity pattern. The algorithm includes a non-linear sigmoidal coherence-to-gain relationship controlled by an online measurement of the present coherence statistics. This processing is particularly effective in acoustic scenarios where the coherence distributions arising from the direct sound and the early reflections are clearly separable from those originating from the diffuse field. In contrast, the effectiveness of the approach presented by Allen et al. [1] is limited in many situations because of the apparent coherence distribution. The performance of the proposed algorithm was subjectively evaluated with ten normal-hearing listeners, assessing the amount of reverberation and overall quality in an experiment inspired by a multiple stimuli with hidden reference test (MUSHRA) [7]. A binaural spectral subtraction method, based on Lebart et al. [2], and a binaural version of the original method described by Allen et al. were used as reference systems. It is demonstrated that the novel coherence-based approach is more successful in acoustic scenarios that exhibit a significant spread in the coherence distribution. Therefore, the proposed algorithm is particularly useful for de-reverberation in large rooms at nearby source-receiver ranges within the reverberation distance.