A Distance-Based Howling Canceller with Adaptive Bandwidth

Public-address system has been widely used in many applications to amplify speech and music signals. In such environments, an acoustic feedback from a loudspeaker to a microphone usually exists and it may causes howling which explosively increases a certain frequency amplitude. The howling is a persistent and an annoying problem in many acoustic enviroments. To solve this problem, we have previously proposed a howling canceller which uses only distance information between the loudspeaker and the microphone [1] while conventional howling cancellers use a complicated feedback path from the loudspeaker to the microphone [2]–[4]. The howling canceller proposed in [1] is based on a property that a howling frequency depends on the distance between the loudspeaker and the microphone. The discance-based howling canceller repeats the distance measurement and frequency removal to deal with environmental changes. The howling canceller does not require feedback path estimation. In the distance-based howling canceller, we have utilized a pilot signal for estimating the distance. Using the estimated distance, we compute multiple frequency candidates of the howling, called howling frequencies, and remove them by cascading notch filters that have nulls at the howling frequencies. We however have a problem about quality degradation of the howling canceller output. In the howling canceller, the notch filters have nulls at howling frequencies estimated from the distance. When the distance becomes long, the width between neighboring nulls of notch filters becomes narrow accordingly. The neighboring bandwidths of notch filters are hence overlapped when the distance is long, and consequently a sound quality of the howling canceller output degrades. In this paper, to solve this problem, we propose an adaptive bandwidth of the notch filter. The adaptive bandwidth narrows when the distance becomes long. We evaluate the speech quality of the howling canceller output with the adaptive bandwidth. We also evaluate its capability of the howling cancellation with the adaptive bandwidth in a practical environment.