注：最近在学习kaldi语音识别工具，在安装过程中遇到了许多问题，为了解决问题，我把ubuntu和一些软件装了又卸，卸了又装，解决了旧问题，又出现新问题，所以在此记录，以备后需。
在一开始，我看了这篇博客（），该博客中的kaldi是在github上下载的，所以要先下载git。
按照步骤进行，可以顺利下载kaldi源码，安装过程也可以按照其中的INSTALL文件进行，
在tools中编译使用到的工具以及在src中配置kaldi的时候，会提示你需要安装一些必需的软件包，按照提示安装即可。
（需要注意的是libatlas3-base软件包无法下载，查询了许多资料以及升级了ubuntu也没办法解决，后来发现有些资料安装的是libatlas-dev和libatlas-base-dev，安装之后发现，上一步骤的编译成功了。但是在随后的编译kaldi的过程中，指令行每次都会停留在同一个界面，刚开始以为是正常过程，但中途大概等了四五个小时之后还是没有进展，索性放弃。如果以后知道问题出在哪里，会尝试继续进行）
然后，我重新安装了kaldi-trunk（kaldi的最新版），进行了安装：
1.需要安装的软件包：gcc,automake,autoconf,subversion,libtool,libatlas-dev,libatlas-base-dev,gfortran
例：sudo apt-get install gcc
2.下载最新版kaldi，即kaldi-trunk
svn co https://svn.code.sf.net/p/kaldi/code/trunk kaldi-trunk
3.编译使用到的tools下的工具(这个过程需要花费一些时间)
cd kaldi-trunk/tools
4.配置kaldi(在kaldi-trunk/src下进行)
./configure
5.最后编译kaldi
然后就可以运行kaldi自带的实例了。按照这样的安装过程应该不会出现什么问题，而且每个步骤是否成功都会有明显的提示，如果没有提示的话，那就需要找找问题所在了。
阅读(...) 评论()steps/make_mfcc_pitch.sh: line 138: run.pl: command not found
ERROR (make-h-transducer[5.4]:GetHmmAsFst():hmm-utils.cc:39) Context size mismatch, ilabel-info [from context FST is 3, context-dependency object expects 1
昨天在kaldi-trunk下面训练模型的时候，一直报各种各样的错。整个人都不好啦，后来师兄觉得可能是kaldi版本的问题，所以让我重新装了kaidi-master.今天重装了，还重新编译了srilm.....然后我直接把训练模型的包直接拖过来运行之后还会一样的错误。我猜想这两个问题都可能是因为我在kaldi-master下面直接把kaldi-trunk下面建立的包拉过来继续训练。这里面用的steps和utils都是kaldi-trunk下面的。所以我在master/egs下面找了一个包进去想拷贝steps和utils来训练模型，结果直接拷贝过去的都会有一把锁，查了网上的用 chmod 777 -R 这个
并不能解决问题。郁闷了一阵儿之后，我尝试在steps原来的包下先右键创建一个链接，再把这个链接复制或者剪切到模型的包里面，在接着运行 run.sh 正在等结果，希望一切顺利。 （后来跟师兄说了这个过程，师兄说这两个包可以直接在wsj下的s5下拷贝，所有有箭头的steps和utils都是指向这里。。嘤嘤嘤）
这样改了以后
，确实没有报那个错误了。
我突发奇想，把master的steps和utils拷贝到trunk里面，再运行run.sh训练了一下模型,竟然发现可以畅通无阻的运行。。。。。。wtf ，困扰了两天的问题，就这么歪打正着。。解决了。喜大普奔。。
需要特别提醒的是，一定要注意各个脚本里面的路径问题。主要是音频文件的路径，还有ngram.sh里面的srilm的路径，注意一下。
希望大家（还有我这个小菜鸡）学习kaldi都能顺利一点。
MFCC一些知识与Kaldi中的MFCC特征解读
因为想用Kaldi去实现一个GMM-UBM的说话人识别和loglike打分，所以想把Kaldi计算的统计量对一对，看能不能把二进制文件里的数值单独拿出来，自己写个代码做结合Kaldi中的MAP做统计量...
Kaldi 中MFCC特征可视化
训练后MFCC文件夹中会产生提取的说话人语音特征，可视化这个文件的命令为：
~/kaldi-master/src/featbin/copy-feats ark:raw_mfcc_pitch_de...
kaldi使用的相关问题和解决
一、编译相关
1.libcublas.so.7.5: cannot open shared object file: No such file or directory
原因：使用的kaldi是从...
如何利用kaldi提自己想要的特征（mfcc plp pitch）
转自：http://blog.csdn.net/wbgxx333/article/details/
基于&em&mfcc&/em&特征的dtw算法实现 立即下载
上传者: ninijungle 时间:
综合评分: 4 积分/C币:3
wav提取&em&mfcc&/em&和&em&pitch&/em& 立即下载
上传者: yangjing_hit...
python提取&em&MFCC&/em& 立即下载
上传者: weixin_ 时间:
综合评分: 1 积分/C币:5
wav提取&em&mfcc&/em&和&em&pitch&/em& 立即下载
上传者: yangjing_hit 时间:...
学习kaldi已经接近两个月了，终于将kaldi中语音的特征数据提取出来，对于其分帧的标签进行对齐，即将输入到TensorFlow搭建的模型中，下面详细的讲解一下如何提取语音的特征以及对音素标签进...
首先，需要完成 标准的 GMM-HMM 声学模型的训练
训练 monophone model 是通过 GMM-HMM System 做 utterance-level t...
我们如何运用已经训练好的模型进行语音识别呢？这才是我们研究的目的啊，是不？
很好，细心的你一定会发现kaldi源码src目录中有online*相关的模块，这就是我们今天的主角啦！！！
Kaldi中...
在这个笔记中，我会首先介绍表示HMM的类HmmTopology和TransitionModel，然后介绍三音素GMM训练脚本train_deltas.sh用到的几个程序，这几个程序与单音素GMM的不同...
没有更多推荐了，Kaldi / Discussion /
Help:Make MFCC features
Discussion
Brought to you by:
Make MFCC features
I am doing aurora4 example from the kaldi-trunk.I am stuck at make MFCC features..when i run
x=train_si84_clean
steps/make_mfcc.sh
data/$x exp/make_mfcc/$x $mfccdir || exit 1;
it outputs
steps/make_mfcc.sh --nj 10 data/train_si84_clean exp/make_mfcc/train_si84_clean mfcc
Successfully validated data-directory data/train_si84_clean
steps/make_mfcc.sh: [info]: no segments file exists: assuming wav.scp indexed by utterance.
run.pl: 10 / 10 failed, log is in exp/make_mfcc/train_si84_clean/make_mfcc_train_si84_clean.*.log
and the log says
Started at Wed May 14 17:56:13 CEST 2014
Invalid option --compress=true
Copy features [and possibly change format]
Usage: copy-feats [options] in-rspecifier out-wspecifier
: Read input as HTK features (bool, default = false)
--sphinx-in
: Read input as Sphinx features (bool, default = false)
Standard options:
: Configuration file with options (string, default = "")
: Print out usage message (bool, default = false)
--print-args
: Print the command line arguments (to stderr) (bool, default = true)
: Verbose level (higher-&more logging) (int, default = 0)
Command line was: copy-feats --compress=true ark:- ark,scp:/home/zamais/kaldi-trunk/egs/aurora4/s5/mfcc/raw_mfcc_train_si84_clean.5.ark,/home/zamais/kaldi-trunk/egs/aurora4/s5/mfcc/raw_mfcc_train_si84_clean.5.scp
compute-mfcc-feats --verbose=2 --config=conf/mfcc.conf scp:exp/make_mfcc/train_si84_clean/wav_train_si84_clean.5.scp ark:-
Kindly help me out.I have not found any relevant information regarding that error..Thanks&
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Possibly you have updated your repository but not re-compiled the source in
a while, or you have an older version of Kaldi on your path.
On Wed, May 14, 2014 at 12:04 PM, Idrees
I am doing aurora4 example from the kaldi-trunk.I am stuck at make MFCC
features..when i run
x=train_si84_clean
steps/make_mfcc.sh --nj 10 \ data/$x exp/make_mfcc/$x $mfccdir || exit 1;
it outputs
steps/make_mfcc.sh --nj 10 data/train_si84_clean
exp/make_mfcc/train_si84_clean mfcc
Successfully validated data-directory data/train_si84_clean
steps/make_mfcc.sh: [info]: no segments file exists: assuming wav.scp
indexed by utterance.
run.pl: 10 / 10 failed, log is in
exp/make_mfcc/train_si84_clean/make_mfcc_train_si84_clean.*.log
and the log says
Started at Wed May 14 17:56:13 CEST 2014
Invalid option --compress=true *
Copy features [and possibly change format]
Usage: copy-feats [options] in-rspecifier out-wspecifier
--htk-in : Read input as HTK features (bool, default = false)
--sphinx-in : Read input as Sphinx features (bool, default = false)
Standard options:
--config : Configuration file with options (string, default = "")
--help : Print out usage message (bool, default = false)
--print-args : Print the command line arguments (to stderr) (bool, default
--verbose : Verbose level (higher-&more logging) (int, default = 0)
Command line was: copy-feats --compress=true ark:-
ark,scp:/home/zamais/kaldi-trunk/egs/aurora4/s5/mfcc/raw_mfcc_train_si84_clean.5.ark,/home/zamais/kaldi-trunk/egs/aurora4/s5/mfcc/raw_mfcc_train_si84_clean.5.scp
compute-mfcc-feats --verbose=2 --config=conf/mfcc.conf
scp:exp/make_mfcc/train_si84_clean/wav_train_si84_clean.5.scp ark:-
Accounting: time=3 threads=1 Ended (code 256) at Wed May 14 17:56:16 CEST
2014, elapsed time 3 seconds
Kindly help me out.I have not found any relevant information regarding
that error..Thanks
Make MFCC features
Sent from sourceforge.net because you indicated interest in
To unsubscribe from further messages, please visit
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks for your response. Actually i have kaldi-stable and kaldi-trunk on my system.I have recompiled kaldi-trunk.Kaldi-trunk did not had copy-feats function.First i gave copy-feats path from kaldi stable.But now i have copied that copy-feats file from kaldi-stable to kaldi-trunk.First it was giving error copy-feats command not found.Now it is not giving that error but the above mentioned.I Don't know what to do next...&
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Probably you have a very old version of kaldi-trunk and
you need to
update: go to the top level and do "svn up".
On Tue, May 20, 2014 at 9:50 AM, Idrees
Thanks for your response. Actually i have kaldi-stable and kaldi-trunk on
my system.I have recompiled kaldi-trunk.Kaldi-trunk did not had copy-feats
function.First i gave copy-feats path from kaldi stable.But now i have
copied that copy-feats file from kaldi-stable to kaldi-trunk.First it was
giving error copy-feats command not found.Now it is not giving that error
but the above mentioned.I Don't know what to do next...
Make MFCC features
Sent from sourceforge.net because you indicated interest in
To unsubscribe from further messages, please visit
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
It is up to date.
At revision 4006.
But still the same :(&
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
That program is in featbin/.
I suspect you did something wrong but without access to your setup I wouldn't be able to help (+ dont' have time anyway)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:&|&&|&&|&&|&&
当前位置： >
语音识别系统之kaldi-----安装出现的错误
作者：u & 来源：转载 &
这几天看到很多人安装kaldi，但是总是在安装的时候一步不能成功。这个原因也比较简单，都是由于自己的linux未配置好。下面我把常见的两个未配置的软件放在这里，也算为安装kaldi做一个归纳。以后kaldi关于安装的问题应该问题不大了。
第一个错误是：whichaclocalmake：***[check_required_programs]error1
&& 这几天看到很多人安装kaldi，但是总是在安装的时候一步不能成功。这个原因也比较简单，都是由于自己的linux未配置好。下面我把常见的两个未配置的软件放在这里，也算为安装kaldi做一个归纳。以后kaldi关于安装的问题应该问题不大了。
&& 第一个错误是：which aclocal make ：***[check_required_programs] error 1
&&&&&&&&&&&&&
&&&&&&&& 这个主要原因是因为没有安装aclocal。大家可以通过安装autoconf来解决。linux里安装就直接：sudo apt-get install autoconf即可。还有automake autotools-dev m4这些你最好都安装了。其实，在安装好linux之后自己需要去配置一些常用的软件，具体哪些，以后我总结好就告诉大家。因为linux对我来说就是个工具，所以我没很深入的去理解它。
&& 第二个错误是:gzfilebuf.h:8:18: fatal error :zlib.h :no such file or directory
&&&&&&&&&&&& &
&&&&&&&&&& 这个主要原因是未安装zlib。大家可以参考这个网址：http://topcat.iteye.com/blog/403277。具体内容：
&&&&&&& zlib官方网站： http://www.zlib.net 上下载源码来安装zlib软件包。
目前最新版本zlib是zlib1.2.3，安装开始； $wget http://www.zlib.net/zlib-1.2.3.tar.gz $tar -xvzf zlib-1.2.3.tar.gz $cd zlib-1.2.3.tar.gz $./configure $make $sudo make install
linux有些软件的安装不能使用apt-get，可以下载软件包，然后去安装，无非就是./configure make make install等等。
&&&&&&&好了，这两个常见错误解决了。下面附上我写的安装的博客：
&&&&&&& 1.语音识别工具箱之kaldi介绍
&&&&&&&&2.语音识别系统之kaldi-----安装续
&&& 最后，有什么问题欢迎和我交流。后期有什么错误，我会及时的加上去。谢谢大家和我交流。
版权所有 IT知识库 CopyRight (C)
IT知识库 IT610.com , All Rights Reserved.Phone: 425 247 4129
(Daniel Povey)
Kaldi's code lives at . To checkout (i.e. clone in the git terminology) the most recent changes, you can use this command git clone https://github.com/kaldi-asr/kaldi or follow the github
and click "Download in zip" on the github page (right hand side of the web page)
To browse the model builds that are available (not many), please click on .
If you have any suggestion of how to improve the site, please .