This recipe implements Streaming Zipformer-Transducer model.
See https://k2-fsa.github.io/icefall/recipes/Streaming-ASR/librispeech/zipformer_transducer.html for detailed tutorials.
./emformer.py and ./train.py are basically the same as ./emformer2.py and ./do_not_use_it_directly.py. The only purpose of ./emformer2.py and ./do_not_use_it_directly.py is for exporting to sherpa-ncnn.