vsooda / mxnet_kaldi

use mxnet and kaldi to train asr model

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

note

This is the speech demo from mxnet, but for some reason, they removed this demo. I had contributed timit recipe for this demo, and make it work quite well.

I will update this demo for the latest kaldi and mxnet. currently, we use a old kaldi. we will upgraded to latest in a few day.

thanks for the original author @pluskid and @yzhang87.

Build Kaldi

Build Kaldi as shared libraties if you have not already done so.

cd kaldi/src
./configure --shared # and other options that you need
make depend
make

Build Python Wrapper

  1. Copy or link the attached python_wrap folder to kaldi/src.
  2. Compile python_wrap/
cd kaldi/src/python_wrap/
make

usage (simple version:timit)

  1. run timit recipe in Kaldi. 2 modify egs/timit/s5/global.cfg
  2. ./run.sh

every thing should be ok.

About

use mxnet and kaldi to train asr model


Languages

Language:Python 90.2%Language:Shell 5.5%Language:C++ 4.0%Language:Makefile 0.2%