Create a subword Lexicon FST for Kaldi

This is the code belonging to the paper Improved subword modeling for WFST-based speech recognition.

For each subword marking style (word boundary marker, left-right marked, left-marked, right-marked) a seperate script exists in local/ that can create a L.fst.

The standard way to use this scripts is:

extra=3
utils/prepare_lang.sh --phone-symbol-table data/lang/phones.txt --num-extra-phone-disambig-syms $extra data/subword_dict "<UNK>" data/subword_lang/local data/subword_lang

dir=data/subword_lang
tmpdir=data/subword_lang/local

# Overwrite L_disambig.fst
common/make_lfst_wb.py $(tail -n$extra $dir/phones/disambig.txt) < $tmpdir/lexiconp_disambig.txt | fstcompile --isymbols=$dir/phones.txt --osymbols=$dir/words.txt --keep_isymbols=false --keep_osymbols=false | fstaddselfloops  $dir/phones/wdisambig_phones.int $dir/phones/wdisambig_words.int | fstarcsort --sort_type=olabel > $dir/L_disambig.fst

For the other scripts (l/r/lr-marked ) the number of extra disambiguation symbols can be reduced to 1

What type of marking style is the best?

This unfortunately depends on your language and dataset. We have seen different optimal values for different datasets and languages.

Limitiations

• The lexicon files are not updated in the lang directory, so lexicon-based alignment of lattices will not work (fix in progress)

• At this moment all pronunciations will have probability 1 (which is common anyway for grapheme-based systems). If custom probabilities are required the local/make_lfst_*.py files should be updated to include them.

Help

Feel free to make an issue or send me an email on peter.smit@aalto.fi if you have trouble getting these scripts to work.