Introduction

This repo contains ConvEmformer transducer models that have been converted to ncnn format.

You can use models from this repo with https://github.com/k2-fsa/sherpa-ncnn for speech recognition. It runs on x86 machines as well as on embedded devices.

Note: It has only 18.6 M parameters.

If you are interested in model training and conversion, please have a look at https://github.com/k2-fsa/icefall/pull/717

The torchscript model can be found at https://huggingface.co/csukuangfj/icefall-asr-wenetspeech-conv-emformer-transducer-stateless-small-2022-12-08