Intel MKL does not seem to be installed.

[George Kam]

Hi there, I've installed every possible dependency, including Intel MKL, however, when I run 

extras/check_dependencies.sh

I get the following output: 

Intel MKL does not seem to be installed.

 ... Download the installer package for your system from: 

 ...   https://software.intel.com/mkl/choose-download

 ... You can also use other matrix algebra libraries. For information, see:

 ...   http://kaldi-asr.org/doc/matrixwrap.html


I have no idea how to fix that. Can you please help me?

[Daniel Povey]

If mkl is really installed, the installation should work even though check_dependencies.sh is not happy.

--
Go to http://kaldi-asr.org/forums.html to find out how to join the kaldi-help group
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/411c31d3-af64-402f-bdc5-2676c75dd732n%40googlegroups.com.

[JJ Kam]

I checked that everything is installed on my Mac, and ran the compilation script even though it complains about Intel MLK not being installed. Here is what I got:

Configuring KALDI to use MKL.

Checking compiler g++ ...

Checking OpenFst library in  ...

***configure failed: Could not find file /include/fst/fst.h:

  you may not have installed OpenFst. See ../tools/INSTALL ***

Makefile:29: kaldi.mk: No such file or directory

kaldi.mk does not exist; you have to run ./configure

make: *** [kaldi.mk] Error 1

Makefile:29: kaldi.mk: No such file or directory

kaldi.mk does not exist; you have to run ./configure

make: *** [kaldi.mk] Error 1

Is it related to that Intel MLK issue?

понедельник, 5 июля 2021 г. в 17:26:21 UTC+3, Dan Povey:

[Daniel Povey]

Unrelated.

Follow the instruxtions it printed.

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/92ccbe88-f26f-4835-a4ab-b71003b571ban%40googlegroups.com.

[软件开发工作经验]

the detail msg is:

steps/online/nnet2/extract_ivectors_online.sh: extracting iVectors

steps/online/nnet2/extract_ivectors_online.sh: extracting iVectors

run.pl: 10 / 10 failed, log is in exp/nnet2_online/ivectors_test_eval92/log/extract_ivectors.*.log

run.pl: 10 / 10 failed, log is in exp/nnet2_online/ivectors_test_dev93/log/extract_ivectors.*.log

run.pl: 10 / 10 failed, log is in exp/nnet2_online/ivectors_test_eval93/log/extract_ivectors.*.log

local/online/run_nnet2_common.sh: error extracting iVectors.

(base) server@server-PowerEdge-R740:~/kaldi-trunk/egs/wsj/s5$ cat exp/nnet2_online/ivectors_test_eval92/log/extract_ivectors.2.log

# ivector-extract-online2 --config=exp/nnet2_online/ivectors_test_eval92/conf/ivector_extractor.conf ark:data/test_eval92_hires/split10/2/spk2utt scp:data/test_eval92_hires/split10/2/feats.scp ark:- | copy-feats --compress=true ark:- ark,scp:/home/server/kaldi-trunk/egs/wsj/s5/exp/nnet2_online/ivectors_test_eval92/ivector_online.2.ark,/home/server/kaldi-trunk/egs/wsj/s5/exp/nnet2_online/ivectors_test_eval92/ivector_online.2.scp

# Started at Wed Jun 30 20:48:10 CST 2021

#

copy-feats --compress=true ark:- ark,scp:/home/server/kaldi-trunk/egs/wsj/s5/exp/nnet2_online/ivectors_test_eval92/ivector_online.2.ark,/home/server/kaldi-trunk/egs/wsj/s5/exp/nnet2_online/ivectors_test_eval92/ivector_online.2.scp

ivector-extract-online2 --config=exp/nnet2_online/ivectors_test_eval92/conf/ivector_extractor.conf ark:data/test_eval92_hires/split10/2/spk2utt scp:data/test_eval92_hires/split10/2/feats.scp ark:-

LOG (ivector-extract-online2[5.5.839~8-0c6a]:ComputeGconsts():diag-gmm.cc:119) num gs256dim40offset-36.7575

LOG (ivector-extract-online2[5.5.839~8-0c6a]:ComputeDerivedVars():ivector-extractor.cc:183) Computing derived variables for iVector extractor

LOG (ivector-extract-online2[5.5.839~8-0c6a]:ComputeDerivedVars():ivector-extractor.cc:204) Done.

ERROR (ivector-extract-online2[5.5.839~8-0c6a]:OnlineTransform():online-feature.cc:533) Dimension mismatch: source features have dimension 280 and LDA #cols is 91

 

[ Stack-Trace: ]

/home/server/kaldi-trunk/src/lib/libkaldi-base.so(kaldi::MessageLogger::LogMessage() const+0xb42) [0x7f6f0bc5d692]

ivector-extract-online2(kaldi::MessageLogger::LogAndThrow::operator=(kaldi::MessageLogger const&)+0x21) [0x55c9bf21f4e9]

/home/server/kaldi-trunk/src/lib/libkaldi-feat.so(kaldi::OnlineTransform::OnlineTransform(kaldi::MatrixBase<float> const&, kaldi::OnlineFeatureInterface*)+0x1b6) [0x7f6f07e255f6]

/home/server/kaldi-trunk/src/lib/libkaldi-online2.so(kaldi::OnlineIvectorFeature::OnlineIvectorFeature(kaldi::OnlineIvectorExtractionInfo const&, kaldi::OnlineFeatureInterface*)+0x1fb) [0x7f6f0c5b4b5d]

ivector-extract-online2(main+0x9dd) [0x55c9bf21d416]

/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xe7) [0x7f6f0ad29bf7]

ivector-extract-online2(_start+0x2a) [0x55c9bf21c92a]

 

kaldi::KaldiFatalErrorLOG (copy-feats[5.5.839~8-0c6a]:main():copy-feats.cc:143) Copied 0 feature matrices.

# Accounting: time=0 threads=1

# Ended (code 1) at Wed Jun 30 20:48:10 CST 2021, elapsed time 0 seconds

I've passed by 20 files,now get error with 9000 files,I've modifed subset line and nj number,still get error.

thanks

[Daniel Povey]

Likely about 13-dim vs 40-dim features

--

Go to http://kaldi-asr.org/forums.html to find out how to join the kaldi-help group
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/tencent_7766ED700B7F7FF9E90DE4B80DE897350105%40qq.com.

[软件开发工作经验]

Is “40-dim” about "make-mfcc.sh -nj 40"?

what is the source of "13-dim"?

  steps/online/nnet2/extract_ivectors_online.sh --cmd "$train_cmd" --nj 2 \

    data/train_si284_hires_max2 exp/nnet2_online/extractor exp/nnet2_online/ivectors_train_si284 || exit 1;

the above line is ok,and next extract_ivectors_online.sh get error,so i'm very confused,i've compared dir in first parameter ,first dir use below line to make.

  steps/online/nnet2/copy_data_dir.sh --utts-per-spk-max 2 data/train_si284 \

    data/train_si284_hires_max2

but the source data is same.

it's so hard,now i try to use nnet3,hope make it pass.

------------------ 原始邮件 ------------------

发件人: "kaldi-help" <dpo...@gmail.com>;

发送时间: 2021年7月12日(星期一) 上午10:48

收件人: "kaldi-help"<kaldi...@googlegroups.com>;

主题: Re: [kaldi-help] wsj's nnet2 error "Dimension mismatch: source features have dimension 280 and LDA #cols is 91"

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/CAEWAuySbVEU6%3D2KN%2B8%2BP_q%2BSNaRvd-5Wen6WF9NhOzd-Uz3cnw%40mail.gmail.com.

[Daniel Povey]

It's not about the number of jobs, it's about using the hires config, conf/mfcc_hires.conf, versus conf/mfcc.conf (or no config)

To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/tencent_30A23699DFC984D33586ADB87DF827579007%40qq.com.