Does Kaldi fail to extract features from the non-standard pronunciation of ethnic minorities?

steps/make_mfcc_pitch.sh --nj 64 --mfcc-config conf/mfcc_hires.conf --cmd run.pl data/test
utils/validate_data_dir.sh: Successfully validated data-directory data/test
steps/make_mfcc_pitch.sh: [info]: no segments file exists: assuming wav.scp indexed by utterance.
steps/make_mfcc_pitch.sh: It seems not all of the feature files were successfully procesed (4379 != 4596); consider using utils/fix_data_dir.sh data/test
steps/make_mfcc_pitch.sh: Succeeded creating MFCC and pitch features for test
steps/compute_cmvn_stats.sh data/test
Succeeded creating CMVN stats for test
fix_data_dir.sh: kept 4379 utterances out of 4596
fix_data_dir.sh: old files are kept in data/test/.backup

Start by running the reported steps/make_mfcc_pitch.sh and steps/compute_cmvn_stats.sh commands with conf/mfcc_hires.conf on data/test, then inspect utils/validate_data_dir.sh and utils/fix_data_dir.sh. Compare the 4,379 processed utterances with the 4,596 input utterances and identify what causes the missing feature files; done means the cause and expected handling are documented or fixed.

Written by the indexing model from the issue text.