Acoustic Howling Suppression with Neural Networks using Real-world Data

Abstract

Due to the closed loop between the microphone and loudspeaker, howling occurs in scenarios such as meetings, KTV, and smart classroom. The howling will not only affect the people's hearing, but also damage audio equipment, bringing poor experiences to listeners. The current high-end devices are equipped with howling suppression algorithms, where traditional signal processing methods such as frequency shift, notch filter, and adaptive feedback cancellation are usually adopted. In this work, we attempt to leverage the power of deep learning techniques to perform howling suppression. To that aim, we set up an environment in which the howling sound is generated and recorded. Based on this setting, we totally collected 36 hours data for the purpose of training. To demonstrate the potentials of deep learning, several networks have been experimented and the corresponding results are reported. The results show that the deep learning-based methods produce a superior howling suppression performance, which indicates the feasibility of using deep learning in this task.