This is a small conv-emformer-transducer model trained on LibriSpeech. Please refer to https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/conv_emformer_transducer_stateless2 for the training recipe.
Please refer to train.sh for model configurations such as num-encoder-layers, memory-size , nhead etc.
After training, the model can be exported to ncnn format using export-ncnn.sh.