This tar file contains perl scripts designed to manipulate speechdatabase transcriptions and word lattice files.
Demonstration Scripts_________________________
demo_lattices.sh - Automatically generate landmark-based pronunciationmodels from the pronlex dictionary. Use these to augment the edgesin a set of Switchboard lattices. Find minimum-distance alignmentof pronunciation models of words on coterminous edges.
demo_praat.sh - Convert TIMIT transcriptions to Praat TextGrid format,by way of HTK-format MLF.
Tool Scripts__________________________________
align_edges.pl - Compute the alignment between pronunciation tagson parallel edges in a word lattice.
augment_lattice.pl - Augment a word lattice with pronunciationsfrom a dictionary.
cfg - contains configuration files for phn2lm.pl.
consecutivelabels.pl - Separate a string of labels into consecutivelines, e.g., in order to divide the lines in a WS97 syllable transcription
dict2mlf.pl - convert a dictionary file into an MLF-format"pseudo-transcription."
extract_examples.pl - extract training examples for a neural networkor support vector machine, from waveform files or MFCC files, basedon transcriptions given in an MLF.
extractmlflayer.pl - Extract one layer from a multi-layer MLF.
lat2praat.pl - Create a Praat transcription file that summarizes some of thealternative segmentations available in an HTK-format word lattice.
lattice_cheat.pl - Compute the distance between the dictionarypronunciation model for each edge in a lattice vs. a known phonemetranscription of the same data. Find the path through the latticethat minimizes this distance.
layermlfs.pl - Construct a multilayer MLF from several single-layerMLFs.
makescripts.pl - Search a directory tree to find examples of fileswith extensions matching a regular expression; list these files inan HTK-format "script" file.
mlf2dict.pl - Create a dictionary file based on the alignment betweenphoneme and word transcriptions in a multilayer MLF.
mlf2praat.pl - Convert an MLF to a Praat-format transcription.
mlf2tgt.pl - Create a binary-valued neural network target file basedon segmentation provided in an MLF.
normalize_phn.pl - Eliminate unwanted phoneme distinctions from anMLF.
normalize_swb_wrd.pl - Eliminate unwanted word distinctions from anMLF encoding Switchboard word transcripts.
phn2lm.pl - Convert phoneme transcription to a distinctive-featurebased landmark transcription.
pinch_lattice.pl - Pinch a word lattice: remove all nodes but thosein the maximum-likelihood path, and make each edge coterminous withone of the edges in the ML path.
praat2mlf.pl - Convert a Praat transcription to MLF.
prune_lattice.pl - Prune a word lattice to eliminate pathssignificantly less probable than the ML path.
rnc2mlf.pl - Convert transcriptions from the Boston Radio News corpusinto MLF.
score_lattice.pl - Compute the 1-best WER of a lattice, and thelattice error rate, by comparing to a known word-leveltranscription.
sort_txgd.pl - Sort tiers in a Praat transcription.
sphere2mlf.pl - Convert SPHERE transcriptions to MLF.
split_lattice.pl - Split a multi-utterance lattice file into manysingle-utterance lattice files.
timit2sphinx.pl - Convert TIMIT phoneme labels into SPHINX phonemelabels.
timit2swb.pl - Convert TIMIT phoneme labels into Switchboard phonemelabels.
xwaves2mlf.pl - Convert an XWaves label file (e.g., as used in theBoston Radio News corpus) into MLF.