视频会议中音频多点处理器的研究

H．323视频会议系统中的多点处理单元（MCU）能在多点会议中提供音频、视频或数据流的集中处理。针对音频信息的处理需求，提出了一种实用的混音处理策略，它具有计算复杂较低、会议重点突出的特点，并且一般情况下不会发生溢出问题。

Audio MP in Video Conference

ITU_T H.323 describes the components for multimedia communication systems in those situations where the underlying transport is a packet_based network.The multipoint control unit (MCU) can provide centralized processing of audio,video,and/or data stream in a multipoint conference.MCU is composed of the multipoint processor (MP) and the multipoint controller (MC).MP takes responsibilities of collecting audio,video,and/or data streams from all the terminals of the multipoint conference,processing all the information in the streams,and sending the processed data to the appointed terminals under the control of MC. In this paper,the authors bring forward some solutions for the request of processing audio stream,and then particularly present a practical policy aiming at the audio signals mixing operation.In the centralized multipoint conference mode,it is necessary to do the audio mixing operation on the speech from all the audio channels. The basic audio mixing technology includes three steps.First,MCU decodes the audio code streams from every audio channel respectively,and gets the sum of all the decoded speech.Second,the target speech corresponding to every terminal is gained after subtracting the source signal from the sum.Lastly,the target speech of every terminal is coded respectively,and transmitted to the specific terminal.So each of the terminals receives the audio signal containing all the signal of other terminals. There are many shortcomings in the method above.First,the more the terminals accessing the videoconference are,the more number of speech Codec used by MCU consequently is.Thus the calculating burden of MCU becomes heavy.Second,it is not necessary to mix all the speech from every audio channel equally.It is difficult for the perceptual ability to distinguish the useful information when the speech signals taken into the audio_mixer are more than 4 channels. Therefore,we design an improved audio_mixer that employs a kind of competitive mechanism.When the number of terminals accessing MCU is more than 4,we select 4 channels with the higher speech energy within fixed time interval and take them into the audio_mixer.The speech signals of other channels are regarded as the background noise after a certain of attenuation.The audio_mixer calculates the energy of speech in a fixed time interval and decides the state of every channel according to their speech energy.The state of every channel is preserved until to the end of the following time interval.