tf_kaldi_io 0.1.0

kaldi-io for Tensorflow

A python package: provide a custom tensorflow dataset for kaldi io

Python is its wrapper, C++ is its backend implemention. It depends on two things:

Through [kaldi-io lib](https://github.com/open-speech/kaldi-io.git), it is able to:
- direct read from kaldi rspecifier(scp, ark, in text or binary, just as [kaldi](https://github.com/kaldi-asr/kaldi))
- support multiple feature transforms:
- delta
- cmvn
- splice
- sampling
- compute fast: kaldi Matrix|Vector with blas math lib is used

Through [tensorflow dataset](https://www.tensorflow.org/guide/datasets), it is able to:
- shuffling
- batching at frame or utt level
- bucketing with input sequence lengths
- and all other tensorflow native dataset manipulations and features (parellel, prefetch, ..)