Add delays to make sure we don't process audio too quickly.

src/pipeline/ExternVocoderStep.cpp

@@ -95,15 +95,19 @@ int ExternVocoderStep::getOutputSampleRate() const
 
 std::shared_ptr<short> ExternVocoderStep::execute(std::shared_ptr<short> inputSamples, int numInputSamples, int* numOutputSamples)
 {
- const int MAX_OUTPUT_SAMPLES = 1024;
-
  *numOutputSamples = 0;
  short* outputSamples = nullptr;
 
  // Write input samples to thread for processing
  if (numInputSamples > 0)
  {
  codec2_fifo_write(inputSampleFifo_, inputSamples.get(), numInputSamples);
+#ifdef _WIN32
+ // XXX: simulate required processing time as actual I/O is handled
+ // by another thread. For some reason this is only needed on Windows.
+ int timeToSleep = (numInputSamples * 1000 / getInputSampleRate()) >> 2;
+ if (timeToSleep > 0) Sleep(timeToSleep);
+#endif // _WIN32
  }
 
  // Read and return output samples from thread.