Anybody know the 'chunk_feature' in master branch means what thing?

Deep Speech
#1

[DeepSpeech/evaluate.py]

Line 242 in 5ff3b31

chunk_features = batch_features[:, i:i + N_STEPS, :]

#2

The inference graph can’t take an entire audio file at once, it only handles 16 time steps at a time, hence the need to split samples into chunks of size 16.

#3

Why we cannot take an entire audio file at once? Does it mean processing too long duration audio, such as an hour?

#4

You can, just not with the inference graph. It’s how we implement streaming inference.