LINGUIST List 14.1593
Wed Jun 4 2003
Software: FSM and GRM libraries: Speech Recognition
Editor for this issue: Tomoko Okuno <tomokolinguistlist.org>
Brian Roark, FSM and GRM libraries: Speech Recognition System
Message 1: FSM and GRM libraries: Speech Recognition System
Date: Tue, 03 Jun 2003 10:28:57 +0000
From: Brian Roark <roarkresearch.att.com>
Subject: FSM and GRM libraries: Speech Recognition System
New versions (4.0) of the AT&T FSM and GRM libraries have been
released, and binaries are available for research use.
The FSM library is a set of general-purpose software tools for
building, combining, optimizing, and searching weighted finite-state
acceptors and transducers. Finite-state transducers are automata for
which each transition has an output label in addition to the more
familiar input label. Weighted acceptors or transducers are acceptors
or transducers in which each transition has a weight as well as the
input or input and output labels.
The original goal of the FSM library was to provide algorithms and
representations for phonetic, lexical, and language-modeling
components of large-vocabulary speech recognition systems. This
imposed the following requirements:
Generality: to support the representation and use of the various
information sources in speech recognition
Modularity: to allow rapid experimentation with different
Efficiency: to support competitive large-vocabulary recognition
using automata of more than 10 million states and transitions.
The GRM Library (Grammar Library) is a set of general-purpose software
tools for constructing, modifying, and compiling grammars. It includes
functions and utilities for:
a. Context-Dependent Rewrite Rules: compilation of weighted
context-dependent rules into weighted finite-state transducers
defined over general semirings.
b. Context-Free Grammars: approximation and compilation of weighted
context-free grammars into weighted automata. Dynamic modification
of compiled context-free grammar automata.
c. Text and Grammar Processing Utilities: general text and grammar
processing utilities (e.g., construction of weighted suffix
automata, computation of the expected counts of sequences appearing
in a weighted automaton, construction of local grammars).
d. Statistical Language Models: creation and modification of
statistical language models derived from text or weighted automata.
The GRM library uses many of the functions and representations of
the FSM library and outputs automata that can be directly manipulated
with the FSM library.