A new weighted audio mixing algorithm for a multipoint processor in a VoIP conferencing system

Audio Conferencing is the one of the main features provided by VoIP telecommunication systems. Along with factors such as background noise, low audio level, delay and packet loss, audio mixing algorithm also contributes noise to the output of a audio conferencing system. True mixing algorithm suffers from the problem of overflow / underflow which leads to addition of noise in the form of clipping. Several researchers have proposed many weighted audio mixing algorithms some of which mitigate this problem and increase the voice quality of the mixer output. But in high background noise levels these algorithms fail to maintain the voice quality and lead to lower mean opinion scores. In this paper we introduce a new weighted audio mixing algorithm with some voice enhancement algorithms such as noise reduction, automatic level control and voice activity detection. This new algorithm calculates the weighted factor based on the root mean square values of the input streams of the participants of the conference. This helps the algorithm to adaptively smoothen the input streams and provide a scaled mixer output which is better in perceived speech quality. Perceptual Evaluation of Speech Quality (PESQ) and Perceived Audio Level (PLL) measures are used to compare the results of this new algorithm with earlier work in different background noise levels. Our experimental results demonstrate better and consistent speech quality by this new algorithm in all background noise levels.