Edinburgh Speech Tools  2.4-release
 All Classes Functions Variables Typedefs Enumerations Enumerator Friends Pages
wfst_build

Table of Contents

Build a weighted finite-state transducer

Synopsis

wfst_build [option] [rulefile0] [rulefile1] ... [-type string] [-determinize ] [-detmin ] [-o ofile] [-otype string] [-heap int] [-q ]

Builds and.or processes weighted finite state transducers (WFSTs) from various input formats. This program accepts descriptions in the following formats and converts them to WFSTs

  • regular expressions
  • regular grammars
  • Koskenniemi/Kay/Kaplan context restriction rules

In addition various operations can be performed on two WFSTs

  • compose: form new WFST feeding output of first WFSTs into second WFSTs.
  • union: form new WFST accepting the language both WFSTs
  • intersect: form new WFST accepting only the language common to both WFSTs
  • concat: form new WFST accepting the language from the concatenation of all strings in the first WFST to all strings in the second.

The newly formed WFSTs can be optionally determinized and minimzed.

The option asis allows a single WFSTs to be loaded and determinized and/or minimized

Options

  • -type: string " {kk}" Input rule type: kk, lts, rg, tl, compose, regex union, intersect, concat, asis
  • -determinize: Determinize WFST before saving it
  • -detmin: Determinize and minimize WFST before saving it
  • -o: ofile Output file for saved WFST (default stdout)
  • -otype: string " {ascii}" Output type, ascii or binary
  • -heap: int " {210000}" Set size of Lisp heap, needed for large rulesets
  • -q: Quiet mode, no summary generated