Could anyone please help me to understand the purpose of adding empty mfcc vectors at end of sample(addZeroMfccWindow) in finishstream.
That’s how the model is trained: https://github.com/mozilla/DeepSpeech/blob/db717d7f7315cc45e80fad55e3597227f47a71a0/training/deepspeech_training/train.py#L57-L74
So the native client is implemented to match it.