A coherence-based noise reduction algorithm for binaural hearing aids

In this study, we present a novel coherence-based noise reduction technique and show how it can be employed in binaural hearing aid instruments in order to suppress any potential noise present inside a realistic low reverberant environment. The technique is based on particular assumptions on the spatial properties of the target and undesired interfering signals and suppresses (coherent) interferences without prior statistical knowledge of the noise environment. The proposed algorithm is simple, easy to implement and has the advantage of high performance in coping with adverse signal conditions such as scenarios in which competing talkers are present. The technique was assessed by measurements with normal-hearing subjects and the processed outputs in each ear showed significant improvements in terms of speech intelligibility (measured by an adaptive speech reception threshold (SRT) sentence test) over the unprocessed signals (baseline). In a mildly reverberant room with T60=200, the average improvement in SRT obtained relative to the baseline was approximately 6.5 dB. In addition, the proposed algorithm was found to yield higher intelligibility and quality than those obtained by a well-established interaural time difference (ITD)-based speech enhancement algorithm. These attractive features make the proposed method a potential candidate for future use in commercial hearing aid and cochlear implant devices.