LanguageModelDecoder

Introduction

This is the repo for langauage model decoder. Codes are based on WeNet and Kaldi.

Dependencies

CMake >= 3.14
gcc >= 10.1
pytorch==1.13.1

Please note that this library uses libtorch 1.13.1. If you have other versions of pytorch installed, you may need to uninstall them first and then install the correct version.

Instructions for build and run the language model decoder

Step 1: install the lanauge model decoder first:

cd runtime/server/x86
python setup.py install

Step 2: download the language model.

Step 3: test if you can run the following code. You need ~40GB of memory to run this code.

import neuralDecoder.utils.lmDecoderUtils as lmDecoderUtils

lm_decoder = lmDecoderUtils.build_lm_decoder(
    'path/to/your/language/model',
)