OpenFst tools

fst_extract_subnet

• Extracts sub-FST from FST given by start and end state

fst_traverser

• General FST traverser

Query-by-Example Spoken Term Detection:

Assuming we have HTK phone lattices for both queries and utterances, we have to follow these steps:

1. Convert lattice to confusion network (sausage)

    scripts/lattice2sausage.sh

2. Convert sausages to FSTs

    scripts/sausage2fst_*.sh

3. Compose query and utterance FSTs

    scripts/compose_fst_cn.sh

4. Convert composition features to recs

    scripts/compose_fst_cn_features_to_recs.sh

5. Merge recs to a single MLF file

    scripts/compose_fst_rec2mlf.sh

Additional scripts are in the ./scripts/ subdir. Some of them is more convenient to copy to your $PATH dir (e.q. $HOME/bin)