kaldi清华模型训练使用online脚本日志

目录名没出现在现状里，用现有目录拷贝重命名． 在线识别要用online
steps/make_mfcc.sh: [info]: no segments file exists: assuming wav.scp indexed by utterance.
You are splitting into too many pieces! [reduce $nj] at utils/split_scp.pl line 209, <I> line 5.

steps/align_fmllr.sh --nj 1 --cmd run.pl data/train_960_30k data/lang exp/tri6b exp/nnet2_online/tri6b_ali_30k
cp: cannot stat 'exp/tri6b/tree': No such file or directory
cp: cannot stat 'exp/tri6b/final.mdl': No such file or directory

 

LOG (gmm-boost-silence[5.4.164~6-13ed]:main():gmm-boost-silence.cc:93) Boosted weights for 2 pdfs, by factor of 1
LOG (gmm-boost-silence[5.4.164~6-13ed]:main():gmm-boost-silence.cc:103) Wrote model to -
apply-cmvn --utt2spk=ark:data/train_960_30k/split1/1/utt2spk scp:data/train_960_30k/split1/1/cmvn.scp scp:data/train_960_30k/split1/1/feats.scp ark:-
WARNING (apply-cmvn[5.4.164~6-13ed]:Open():util/kaldi-table-inl.h:106) Failed to open script file data/train_960_30k/split1/1/feats.scp
ERROR (apply-cmvn[5.4.164~6-13ed]:SequentialTableReader():util/kaldi-table-inl.h:860) Error constructing TableReader: rspecifier is scp:data/train_960_30k/split1/1/feats.scp

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*)
kaldi::MessageLogger::~MessageLogger()
kaldi::SequentialTableReader<kaldi::KaldiObjectHolder<kaldi::Matrix<float> > >::SequentialTableReader(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)
main
__libc_start_main
_start

splice-feats --left-context=3 --right-context=3 ark:- ark:-
transform-feats exp/tri4b/final.mat ark:- ark:-
LOG (transform-feats[5.4.164~6-13ed]:main():transform-feats.cc:161) Applied transform to 0 utterances; 0 had errors.
WARNING (gmm-align-compiled[5.4.164~6-13ed]:main():gmm-align-compiled.cc:103) No features for utterance A01_002
WARNING (gmm-align-compiled[5.4.164~6-13ed]:main():gmm-align-compiled.cc:103) No features for utterance A01_003
WARNING (gmm-align-compiled[5.4.164~6-13ed]:main():gmm-align-compiled.cc:103) No features for utterance A01_005
LOG (gmm-align-compiled[5.4.164~6-13ed]:main():gmm-align-compiled.cc:135) Overall log-likelihood per frame is -nan over 0 frames.
LOG (gmm-align-compiled[5.4.164~6-13ed]:main():gmm-align-compiled.cc:137) Retried 0 out of 3 utterances.
LOG (gmm-align-compiled[5.4.164~6-13ed]:main():gmm-align-compiled.cc:139) Done 0, errors on 3
WARNING (gmm-align-compiled[5.4.164~6-13ed]:Close():kaldi-io.cc:515) Pipe apply-cmvn  --utt2spk=ark:data/train_960_30k/split1/1/utt2spk scp:data/train_960_30k/split1/1/cmvn.scp scp:data/train_960_30k/split1/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri4b/final.mat ark:- ark:- | had nonzero return status 256
# Accounting: time=0 threads=1

 

WARNING (apply-cmvn[5.4.164~6-13ed]:ReadScriptFile():kaldi-table.cc:34) Error opening script file: data/train_960_30k/split1/1/cmvn.scp
ERROR (apply-cmvn[5.4.164~6-13ed]:RandomAccessTableReader():util/kaldi-table-inl.h:2512) Error opening RandomAccessTableReader object  (rspecifier is: scp:data/train_960_30k/split1/1/cmvn.scp)

 

apply-cmvn --utt2spk=ark:data/train_960_30k/split1/1/utt2spk scp:data/train_960_30k/split1/1/cmvn.scp scp:data/train_960_30k/split1/1/feats.scp ark:-
splice-feats --left-context=3 --right-context=3 ark:- ark:-
ERROR (apply-cmvn[5.4.164~6-13ed]:HasKey():util/kaldi-table-inl.h:2639) Attempting to read key A01_001, which is not present in utt2spk map or similar map being read from ark:data/train_960_30k/split1/1/utt2spk

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMessageEnvelope const&, char const*)
kaldi::MessageLogger::~MessageLogger()
kaldi::RandomAccessTableReaderMapped<kaldi::KaldiObjectHolder<kaldi::Matrix<double> > >::HasKey(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)
main
__libc_start_main
_start

LOG (transform-feats[5.4.164~6-13ed]:main():transform-feats.cc:161) Applied transform to 0 utterances; 0 had errors.

[ Stack-Trace: ]

kaldi::MessageLogger::HandleMessage(kaldi::LogMe

align_fmllr.sh执行报下面错误

steps/align_fmllr.sh: feature type is lda
steps/align_fmllr.sh: compiling training graphs
steps/align_fmllr.sh: aligning data in data/train_960_30k using exp/tri4b/final.alimdl and speaker-independent features.
run.pl: job failed, log is in exp/nnet2_online/tri6b_ali_30k/log/align_pass1.1.log

Pipe apply-cmvn  --utt2spk=ark:data/train_960_30k/split1/1/utt2spk scp:data/train_960_30k/split1/1/cmvn.scp scp:data/train_960_30k/split1/1/feats.scp ark:- | splice-feats --left-context=3 --right-context=3 ark:- ark:- | transform-feats exp/tri4b/final.mat ark:- ark:- | had nonzero return status 256

 

 

kaldi中现有的cmvn处理包含三种形式，即apply-cmvn、apply-cmvn-online、apply-cmvn-slide，这三种形式在使用时略有区别。
cmvn为倒谱均值方差归一化，大多数情况下只会对均值进行归一化，也可以写成cmn。
倒谱均值方差归一化,顾名思义，需要先通过compute-mfcc计算倒谱特征mfcc，然后通过compute-cmvn-stats计算相关的cmvn统计量
apply-cmvn归一化时，--utt2spk作为可选参数，该参数决定是对每个句子按照说话人归一化还是整体归一化，当utt2spk作为输入时，此时会分别统计每个说话人对应句子的归一化参数

• 然后对于输入句子的每一帧，将每一帧的各个维度利用上述统计的归一化参数进行归一化