# Speech recognition with Next-gen Kaldi

This repo contains [ncnn][ncnn] model files, which are converted from
models trained using <https://github.com/k2-fsa/icefall/pull/558>
via [pnnx][pnnx].

The inference framework is <https://github.com/k2-fsa/sherpa-ncnn>

# Usage

```bash
sudo apt-get install git-lfs

git clone https://github.com/k2-fsa/sherpa-ncnn
cd sherpa-ncnn
mkdir build
cd build
cmake ..
make -j6
cd ..

# Now download the pretrained model

git lfs install
git clone https://huggingface.co/csukuangfj/sherpa-ncnn-2022-09-05

./build/bin/sherpa-ncnn \
  ./sherpa-ncnn-2022-09-05/tokens.txt \
  ./sherpa-ncnn-2022-09-05/bar/encoder_jit_trace-iter-468000-avg-16-pnnx.ncnn.param \
  ./sherpa-ncnn-2022-09-05/bar/encoder_jit_trace-iter-468000-avg-16-pnnx.ncnn.bin \
  ./sherpa-ncnn-2022-09-05/bar/decoder_jit_trace-iter-468000-avg-16-pnnx.ncnn.param \
  ./sherpa-ncnn-2022-09-05/bar/decoder_jit_trace-iter-468000-avg-16-pnnx.ncnn.bin \
  ./sherpa-ncnn-2022-09-05/bar/joiner_jit_trace-iter-468000-avg-16-pnnx.ncnn.param \
  ./sherpa-ncnn-2022-09-05/bar/joiner_jit_trace-iter-468000-avg-16-pnnx.ncnn.bin \
  ./sherpa-ncnn-2022-09-05/test_wavs/1089-134686-0001.wav
```

If you have any issues, please ask at
<https://github.com/k2-fsa/sherpa-ncnn/issues>

Please see
<https://k2-fsa.github.io/icefall/recipes/librispeech/lstm_pruned_stateless_transducer.html>
for how the model is trained and exported to ncnn if you are interested.

[pnnx]: https://github.com/Tencent/ncnn/tree/master/tools/pnnx
[ncnn]: https://github.com/tencent/ncnn
[icefall]: https://github.com/k2-fsa/icefall