Dereverberation and residual echo suppression in noisy environments

Hands-free devices, such as mobile phones, are often used in noisy and reverberant environments. Therefore, the received microphone signal contains not only the desired speech (commonly called near-end speech) signal, but also interferences such as reverberations of the desired source, background noise, and a far-end echo signal that results from a sound that is produced by the loudspeaker. These interferences degrade the fidelity and intelligibility of the near-end speech and decrease the performance of automatic speech recognition systems. Acoustic echo cancellers are widely used to cancel the far-end echo. Postprocessors, employed in conjunction with acoustic echo cancellers, further enhance the near-end speech. Most post-processors that are described in the literature only suppress background noise and residual echo, i.e., echo which is not suppressed by the acoustic echo canceller. The intelligibility of the nearend speech also depends on the amount of reverberation. Dereverberation techniques have been developed to cancel or suppress reverberation. Recently, practically feasible spectral enhancement techniques to suppress reverberation have emerged that can be incorporated into the post-processor. After a short introduction, the problems encountered in a hands-free device are formulated. A general purpose post-filter is developed, which can be used to suppress non-stationary, as well as stationary, interferences. The problem of dereverberation of noisy speech signals is addressed by using the general purpose post-filter employed to suppress reverberation and background noise. Next, suppression of residual echo is discussed. Finally, a post-processor is developed for the joint suppression of reverberation, residual echo, and background noise. An experimental study demonstrates the beneficial use of the proposed post-processor for jointly reducing reverberation, residual echo, and background noise.