espnet安装--kaldi安装

一、Ubuntu镜像上传到集群
1.下载镜像
代码如下（示例）：
 docker pull ecpe4s/ubuntu20.04-spack

2.查看镜像是否下载完成
代码如下（示例）：
docker images

3.更改镜像名称
代码如下（示例）：
docker tag tensorflow/ecpe4s/ubuntu20.04-spack 211.82.236.15:8888/liu2yang/ubuntu20.04-spack
4.登录私有仓库
代码如下（示例）：

(base) zz@ZZ:~/dockerfile$ docker login 211.82.236.15:8888
1
5.推送镜像
代码如下（示例）：
$ docker push 211.82.236.15:8888/liu2yang/ubuntu20.04-spack

二、ununtu20.04安装CUDA10.0(kaldi基本的深度模型需要CUDA<=10)
参考https://blog.csdn.net/zhiman_zhong/article/details/91491780
      https://blog.csdn.net/ashome123/article/details/105822040
0、下载cuda
从https://developer.nvidia.com/cuda-toolkit-archive下载对应版本的cuda，选择18.04的版本即可。
1、gcc降级
Ubuntu20.04自带的gcc版本为9.7.0，需要添加gcc7才可安装cuda10.2，输入命令安装gcc7
apt-get install gcc-7 g++-7
使用update-alternatives进行版本切换，输入以下命令：
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-7 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 5
2、安装cuda
输入命令进行安装
cd /usr/local/
 cuda_10.2.89_440.33.01_linux.run出错

root@worker01:/usr/local# ls
bin  cuda  cuda-10.0  cuda_10.0.130_410.48_linux.run  etc  games  include  lib  man  pai  sbin  share  src

在主目录下的~/.bashrc文件添加如下路径
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.0/lib64
export PATH=$PATH:/usr/local/cuda-10.0/bin
export CUDA_HOME=$CUDA_HOME:/usr/local/cuda-10.0

root@worker01:/usr/local#  source ~/.bashrc                  
root@worker01:/usr/local# cd /usr/local/cuda/samples/1_Utilities/deviceQuery
root@worker01:/usr/local/cuda/samples/1_Utilities/deviceQuery# ls
Makefile  NsightEclipse.xml  deviceQuery.cpp  readme.txt
root@worker01:/usr/local/cuda/samples/1_Utilities/deviceQuery# make -j 4
root@worker01:/usr/local# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:01_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130

三、安装kaidi(过程参考kaldi语音识别实践P21)

1、下载Kaldi

git clone https://github.com/kaldi-asr/kaldi.git

2、安装依赖库

cd kaldi

tool/extras/check_dependencies.sh

make

3. tool 编译

1)进入下载后的文件目录

cd kaldi/tools/

2)运行编译指令

make

出现错误：gcc: error trying to exec 'cc1plus': execvp: no such file or directory

解决办法：参考https://www.cnblogs.com/mydebug/p/5801117.html
root@worker01:/usr/bin# cd gcc
-bash: cd: gcc: Not a directory
root@worker01:/usr/bin# vim gcc
root@worker01:/usr/bin# ln -sf g++-7 g++
root@worker01:/usr/bin# sudo ln -sf g++-7 x86_64-linux-gnu-g++
root@worker01:/usr/bin# ln -sf gcc-7 gcc
root@worker01:/usr/bin# ln -sf gcov-7 gcov
再编译configure成功

在在主目录下的~/.bashrc文件添加如下路径方便使用指令
export PATH=$PATH:/home/kaldi-master/src/bin/:/home/kaldi-master/src/featbin/:/home/kaldi-master/src/chainbin/:/home/kaldi-master/src/cudafeatbin/:/home/kaldi-master/src/fgmmbin/:/home/kaldi-master/src/fstbin/:/home/kaldi-master/src/gmmbin/:/home/kaldi-master/src/ivectorbin/:/home/kaldi-master/src/kwsbin/:/home/kaldi-master/src/latbin/:/home/kaldi-master/src/lmbin/:/home/kaldi-master/src/nnet2bin/:/home/kaldi-master/src/nnetbin/:/home/kaldi-master/src/nnet3bin/:/home/kaldi-master/src/onlinebin/:/home/kaldi-master/src/online2bin/:/home/kaldi-master/src/rnnlmbin/:/home/kaldi-master/src/sgmm2bin/:/home/kaldi-master/src/tfrnnlmbin