srilm

将srilm-1.7.2解压到tools目录下面，重命名为srilm

1.修改srilm目录下Makefile文件

找到此行： # SRILM = /home/speech/stolcke/project/srilm/devel，另起一行输入srilm的安装路径，SRILM =  （自己的路径）

2.编译srilm

       切换到srilm目录下，执行

　　make World

3.测试

         make test

         

出现如下信息：

        .........

       *** Running test ngram-prune ***

11.17user 2.35system 0:14.20elapsed 95%CPU (0avgtext+0avgdata 0maxresident)k

0inputs+15816outputs (0major+31653minor)pagefaults 0swaps

ngram-prune: stdout output IDENTICAL.

ngram-prune: stderr output IDENTICAL.

*** Running test ngram-server ***

0.21user 0.72system 0:06.56elapsed 14%CPU (0avgtext+0avgdata 0maxresident)k

0inputs+8outputs (0major+2559minor)pagefaults 0swaps

ngram-server: stdout output IDENTICAL.

ngram-server: stderr output IDENTICAL.

*** Running test ppl-counts ***

0.10user 0.24system 0:00.32elapsed 104%CPU (0avgtext+0avgdata 0maxresident)k

0inputs+16outputs (0major+4724minor)pagefaults 0swaps

ppl-counts: stdout output IDENTICAL.

ppl-counts: stderr output IDENTICAL.

*** Running test tagged-ngram ***

0.00user 0.05system 0:00.11elapsed 51%CPU (0avgtext+0avgdata 0maxresident)k

0inputs+24outputs (0major+1789minor)pagefaults 0swaps

tagged-ngram: stdout output IDENTICAL.

tagged-ngram: stderr output IDENTICAL.

*** Running test vocab-aliases ***

0.27user 0.24system 0:01.65elapsed 31%CPU (0avgtext+0avgdata 0maxresident)k

0inputs+16outputs (0major+5845minor)pagefaults 0swaps

vocab-aliases: stdout output IDENTICAL.

vocab-aliases: stderr output IDENTICAL.

需要等待一段时间，如果出现多是IDENTICAL,很少的DIFFERS，就证明srilm编译成功了！

4.先在 .bashrc下添加路径

export PATH="/home/xxt/kaldi/tools/srilm/bin/i686-m64:$PATH"
export PATH="/home/xxt/kaldi/tools/srilm/bin:$PATH"
就可以在任意路径下计算语言模型

ngram-count -text zangzichunwenben -order 3 -write 1.count
ngram-count -read 1.count -order 3 -lm word.3gram.lm -interpolate -kndiscount

或

ngram-count -read 1.count -order 3 -lm word.3gram.lm -interpolate -wbdiscount