Real-Time Howling Detection for Hands-Free Video Conferencing System

This paper presents howling detection method in two -way hands-free communication system. Though hands -free communication is convenient for multi -user system such as audio/video conference, it is very common for the open microphones and loudspeakers produc e acoustic feedback in a closed loop, which results in howling. This obviously prevents any useful conversation between participants. In this paper we propose a real time howling detection method based on the long -term average spectral power and howling information of the previous frame. Performance tests show that the proposed algorithm provides quick and stable howling detection results. Keywords-component; howling detecion; howling suppression; hands -free video conferencing I. INTRODUCTION As high speed internet connectivity has become more easily available, video conferencing is being used more and more everyday life for business and educational purpose, not just for the personal use. Recently, many educational institutions including universiti es are interested in distance education by using the video conferencing technology for students who are separated by time and distance. In the fields of business, various types of video conferencing system has been adopted, because it enable individuals i n distance locations to participate in meetings with time and money saving. In video conferencing system, hands-free communication is more convenient for users. But it is very common to produce acoustic feedback in a closed loop, resulting in howling. Thi s obviously prevents any useful conversation between participants. A typical closed loop path (dashed line) can exist when both parties are using hands-free video conferencing system is illustrated in figure 1. This type of closed loop which includes microphone, network, and loudspeaker is most common in real time hands -free communication system. If the loop gain is greater than unity at one or more frequencies the system becomes unstable and produces oscillations. If the oscillation signal has the freque ncy in the audible range of human hearing, it is called howling or squealing. It not only disturbs normal communications, but also damages power amplifier for overload. In two-way hands-free communication, acoustic feedbacks from loudspeaker into the m icrophone are traditionally cancelled using acoustic echo canceller (AEC). When it has reached enough convergence, it provides protection against howling because it reduces acoustic feedback. Until then, or when the acoustic path is changed or high loudness is required, the system exposed to howling because of improper function of AEC(1,2). One of This research was funded by the MSIP (Ministry of Science, ICT & Future Planning), Korea in the ICT R&D Program 2013. The most popular techniques for supplementing the AEC to prevent howling in hands-free communication systems is notch filter based howling suppression (NHS) methods. NHS methods consist of howling detection and howling suppression by using a notch filter. The design of notch filters for howling su ppression is based on well-established filter design techniques. Hence, it is critical to detect howling signal rapidly and exactly in NHS methods. In this paper we propose a howling detection method