使用matlab自带的VAD以及SCOPE函数进行伪实时的VAD判断，为什么说是伪实时的呢？因为本程序实际上还是使用录音-处理-录音-处理的节奏进行VAD的处理，为了平衡VAD的耗时以及录音时长的冲突，将时间块分的很大（1s），也就是说在更新VAD结果的频率上选择的是不高于1Hz。 废话不多说，实现看以下代码： clc clear all close all time = 10; timeDiv = time / 10 * 0.5 ; timeDiv = max(timeDiv,1); audioRecorder = audiorecorder( 8000, 16 ,1); theSound = zeros(7937,time); theSoundFinal = zeros(1,7936*time)'; dataStart = 1; dataStop = 0; timeCost = 0; %% 对示波器进行初始化 scope = dsp.TimeScope(2,... 'SampleRate', [8000/80 8000], ... 'BufferLength', time * 8000, ... 'YLimits', [-0.3 1.1], ... 'TimeSpan',time,... 'ShowGrid', true, ... 'Title','Decision speech and speech data', ... 'TimeSpanOverrunAction','Scroll',... 'ReduceUpdates',false); %% 对VAD的配置进行初始化 VAD_cst_param = vadInitCstParams; %% 开始录音并处理数据 for recordTime = 1:ceil(time/timeDiv) % 获取当前录音机的状态，如果是关闭的则打开，如果已经打开就直接进行下一步 ifRunning = get(audioRecorder,'Running'); status = strncmp(ifRunning,'on',2); if(status == 0) record(audioRecorder); tic; end timeCost = toc; timeCost = min(timeCost,timeDiv); pause(timeDiv - timeCost) stop(audioRecorder); audioDiv = getaudiodata(audioRecorder)/2; record(audioRecorder); % 重新打开录音机，并开始计时 tic; audioSource = dsp.SignalSource(audioDiv,80); dataLen = length(audioDiv); numTSteps =ceil( dataLen/8000 * 100); while(numTSteps) % 从audioSource（）中提取10ms的数据 speech = audioSource(); % 调用VAD算法 decision = vadG729(speech, VAD_cst_param); % 绘制音频数据波形以及VAD判断结果 scope(decision, speech); numTSteps = numTSteps - 1; end dataStop = dataStop + dataLen; theSoundFinal(dataStart:dataStop,1) = audioDiv; dataStart = dataStop + 1; recordTime; end plot(theSoundFinal(:,1)); ylim([-1 1]); sound(theSoundFinal,8000) pause(time); 运行结果如下：