Last touch to deepspeech 2.0 With iPhone DSP modeled right for inference

Time to do the final run of the training set, with stereo inputs 44KHz and wav files matching the audio characteristics of the iPhone array of microphones, that should give very high accuracy.