Issue in utterancesourcemulti.h of Kaldi2Reader

Jun 23, 2015 at 6:57 AM
Hi,

When I was trying to feed two feature streams as input, it was giving core dumped error. The error in the log file looks like this:

cn.exe: DataReader/Kaldi2Reader/utterancesourcemulti.h:453: msra::dbn::minibatchutterancesourcemulti::minibatchutterancesourcemulti(std::vector<msra::asr::FeatureSection*>&, const std::vector<std::vector<std::basic_string<wchar_t> > >&, const std::vector<std::map<std::basic_string<wchar_t>, std::vector<msra::asr::htkmlfentry> > >&, std::vector< long unsigned int>, std::vector<long unsigned int>, std::vector<long unsigned int>, std::vector<long unsigned int>, size_t, const msra::dbn::latticesource&, const std::map<s td::basic_string<wchar_t>, msra::lattices::lattice::htkmlfwordsequence>&, bool): Assertion `uttframes==framesaccum[i]' failed.

In the training set, some of the utterances do not have alignments. In the header file I printed the values of "uttframes" and "framesaccum[i]". It was giving same values untill the label missing ocured. After the label missing, value of uttframes is giving the value of the number of frames of next utterance of the label missing utterance but the value of framesaccum[i] is becoming next to next utterance frame number of the label missing utterance.
Coordinator
Jul 21, 2015 at 12:38 AM
this should have been fixed in the mater branch.