Low-complexity dereverberation for hands-free audio conferencing unit

This paper proposes a dereverberation method that delivers clear sound even in reverberant environments and discusses its implementation in a hands-free audio conferencing prototype unit. The proposed method is a costeffective single-channel algorithm; it estimates a power spectrum of reverberation based on a linear regression model in the power spectral domain and suppresses the reverberation components using spectral subtraction. The entire calculation complexity of the hands-free prototype unit is reduced for real-time implementation on a low-cost processor. Experiments were conducted to examine the performance of the proposed method. The results indicated that it can achieve almost the same dereverberation performance as that of the conventional method despite its lower computational complexity compared to the conventional method.