Unfortunately that Tensorflow integration didn't include acoustic modeling, so you still need to use Kaldi's neural net toolkit for that.