Adaptive Feedback Canceller with Howling Detection for Hearing Aids

Howling is a very annoying problem to hearing aid users because it restricts the maximum gain of a desired signal to the hearing aid users. In general, howling can be eliminated by using an adaptive feedback canceller which estimates the feedback path from the loudspeaker to the microphone equiped in the hearing aid. When the desired signal is a correlation signal, the feedback canceller often degrades the speech quality which is called as entrainment. To prevent the generation of the entrainment, an adaptive feedback canceller with prediction error filter has been proposed. Since the prediction error filter removes the correlation components from the desired signal, this feedback canceller works for the uncorrelated signal and the entrainment does not occur. Unfortunately, the prediction error filter gives only zeros when the desired signal is a periodic signal. In this case, howling cannot be eliminated because howling is also periodic. In this paper, we propose a feedback canceller that switches two conventional systems, where one eliminates howling and the other eliminates entrainment. The key technology of the proposed method is howling detection which utilizes saturation characteristics of the observed signal. Simulation results show that the proposed method works effectively to cancel both howling and entrainment in comparison to the conventional methods.