tutorial.md

About

This is a Mandarin language ASR system trained on 1505 hours Chinese Mandarin corpus released by DataTang (aidatatang_1505zh). The language model was trained from a large number of colloquial texts.

The aidatatang_1505zh corpus can be free download from here.

Turorial

1. Install Kaldi, refer to the steps here.

2. Download and untar DataTang TDNN Chain Model.

3. Move the untarred directory to kaldi/egs.

4. Go to kaldi/egs/aidatatang_asr, create symlinks to /steps/ and /utils/ like this:

   ln -s ../wsj/s5/steps/ steps
   ln -s ../wsj/s5/utils/ utils

5. Download the aidatatang_1505zh or set your own corpus to WAVpath.

6. Run the model like this:

   ./run.sh WAVpath

7. Check the recognition results like this:

   vi exp/chain/tdnn_1a_sp/decode_offline_test_*/rec_*.txt

   There are samples of the recognition results:

   T0055G0036S0005 来 一部 好看 的 电影
   T0055G0036S0006 我要 拍摄 二维码 啊
   T0055G0036S0007 播放 原来 你 什么 都 不 想要
   T0055G0036S0008 一月 十一日 上午 十八点 提醒 我
   T0055G0036S0009 你 不是 要 给 我 讲个 笑话 吗
   T0055G0036S0010 家 里面 有 一个 小白兔