Acoustic feedback reduction based on Filtered-X LMS and Normalized Filtered-X LMS algorithms in digital hearing aids based on WOLA filterbank

Abstract

The speech signal corrupted by the acoustic feedback in digital hearing aids can be restored by a feedback reduction system using adaptive algorithms such as the least-mean square (LMS) algorithm. The main disadvantage of the LMS algorithm is the instability. In order to avoid this situation, it is used another feedback reduction systems based on two different algorithms: the filtered-X LMS (FXLMS) and the normalized filtered-X LMS (NFXLMS). These algorithms are tested in two digital hearing aid categories: the in-the-ear (ITE) and the in-the-canal (ITC). For both categories, the added stable gain (ASG) value over the limit gain of the digital hearing aids is obtained. The ASG value is achieved as a tradeoff between the segmented signal-to-noise ratio (objective parameter) and the speech quality (subjective parameter). The results show how the digital hearing aid working with a feedback reduction adaptive filter adapted with the NFXLMS algorithm is able to achieve up to 18 dB of increase over the limit gain.