!7

// Note: This is dependent on input buffer size which is not optimal.
// Tested with buffer size 232 ms (using CD quality)

cmodule VAD > DynPrintLogAndEnabled {
  double windowSize = 1; // milliseconds
  transient short[] buffer;
  switchable double threshold =
    // 10.0; // the magic value for 44 KHz stereo
    1; // for 8 KHz mono

  start {
    printLogUpdateInterval = 100;
    // technically we are treating stereo input as mono samples
    // but it might not matter
    dm_addAudioListener(voidfunc(short[] data) enter {
      if (!enabled) ret;
      
      buffer = cloneArray(data);
      dm_q(r {
        if (q.size() > 2) ret with print("overload");
        short[] buf = buffer;
        float sampleRate = dm_audioInputSampleRate();
        if (sampleRate == 0) ret;
        float originalSampleRate = sampleRate;
        if (dm_audioInputChannels() > 1) buf = mixStereoShortArrayToMono(buf);
        buf = convertSampleRate_shortArray_simple(buf, sampleRate, 8000);
        sampleRate = 8000f;
        long time = sysNow();
        int windowLength = msToSamples_int(windowSize, sampleRate);
        //printWithMilliseconds("Window length (" + sampleRate + "): " + windowLength);
        new L<Double> l;
        for (IntRange r : intRangeChunks(buf, windowLength))
          l.add(normalizedAutocorrelationOfAudioSamples(buf, r));
        double result = doubleAverage(l);
        double confidence = result/threshold*50;
        print("VAD " + (confidence >= 50 ? "XXX" : "   ") + " value: " + iround(confidence) + " % [" + elapsedMS(time) + " ms, "
          + nWindows(l(buf)/windowLength) + " of size " + windowLength + ", converted from \*originalSampleRate*/ Hz]");
      });
    });
  }
}