void instruction(int begin, int end, bool needPredict) // [begin, end)
{
int len = end - begin;
for (int i = 0; i < len; i++)
{
CXYNN.SetValue(i, (Int16)samples[begin + i]);
}
CXYNN.AddFrame();
PosteriorHandler.updateWindow();
if (needPredict)
{
int ret = PosteriorHandler.Predict();
double conf = PosteriorHandler.CalcConfident();
Debug.Log(ret + " <- " + conf); // for test
Loger.Log(ret + " <- " + conf); // for test
if (conf >= 0.3)
{
Debug.LogError(ret);
Loger.LogError(ret);
PredictPool.AddPredict(ret);
}
}
}
public static void updateWindow()
{
CXYNN.Predict();
double[] tmp = new double[caseNumber];
CXYNN.calcProb();
for (int i = 0; i < caseNumber; i++)
{
tmp[i] = CXYNN.readProb(i + 1);
}
addFrame(tmp);
}
void Instruction()
{
// 计算每一帧所包含的sample的数量
int bufferLen = frameLength * 16000 / 1000;
// 等待输入足够的音频片段
while (samples.Count < 15 * 16 && syncDataWithClip())
{
}
if (samples.Count < 15 * 16)
{
Debug.Log("do not have enough info, exit");
Loger.Log("do not have enough info, exit");
return;
}
// Mfcc 填充 <= 我们使用的Mfcc代码中需要预先进行填充, 以保证最初的几帧得到的数据是有效的
for (int i = 0; i < 15 * 16; i++)
{
CXYNN.SetPrev(i, (Int16)samples[i]);
}
// 识别过程
int cnt = 0;
int cutFlag = 0;
for (int i = 15 * 16; syncDataWithClip() || i + bufferLen - 1 < samples.Count;)
{
if (i + bufferLen > samples.Count)
{
continue;
}
cnt += 1;
instruction(i, i + bufferLen, cnt >= 30 && cutFlag % 3 == 0);
i += bufferLen;
cutFlag++;
}
// for debug
Debug.Log("Done");
Loger.Log("Done");
}
public void Init()
{
CXYNN.InitCXYNN();
CXYNN.InitMfcc(frameLength * 16000 / 1000);
}