Network optimizations for large-vocabulary speech recognition

Mehryar Mohri, Michael Riley

Research output: Contribution to journalArticle

Abstract

The redundancy and the size of networks in large-vocabulary speech recognition systems can have a critical effect on their overall performance. We describe the use of two new algorithms: weighted determinization and minimization. These algorithms transform recognition labeled networks into equivalent ones that require much less time and space in large-vocabulary speech recognition. They are both optimal: weighted determinization eliminates the number of alternatives at each state to the minimum, and weighted minimization reduces the size of deterministic networks to the smallest possible number of states and transitions. These algorithms generalize classical automata determinization and minimization to deal properly with the probabilities of alternative hypotheses and with the relationships between units (distributions, phones, words) at different levels in the recognition system. We illustrate their use in several applications, and report the results of our experiments.

Original languageEnglish (US)
Pages (from-to)1-12
Number of pages12
JournalSpeech Communication
Volume28
Issue number1
DOIs
StatePublished - May 1999

Fingerprint

Network Optimization
Vocabulary
Speech Recognition
Speech recognition
vocabulary
Alternatives
redundancy
Redundancy
Automata
Eliminate
Transform
Generalise
Unit
experiment
performance
Experiment
Experiments

ASJC Scopus subject areas

  • Signal Processing
  • Electrical and Electronic Engineering
  • Experimental and Cognitive Psychology
  • Linguistics and Language

Cite this

Network optimizations for large-vocabulary speech recognition. / Mohri, Mehryar; Riley, Michael.

In: Speech Communication, Vol. 28, No. 1, 05.1999, p. 1-12.

Research output: Contribution to journalArticle

@article{6fb0449b53324e07895950e57896e781,
title = "Network optimizations for large-vocabulary speech recognition",
abstract = "The redundancy and the size of networks in large-vocabulary speech recognition systems can have a critical effect on their overall performance. We describe the use of two new algorithms: weighted determinization and minimization. These algorithms transform recognition labeled networks into equivalent ones that require much less time and space in large-vocabulary speech recognition. They are both optimal: weighted determinization eliminates the number of alternatives at each state to the minimum, and weighted minimization reduces the size of deterministic networks to the smallest possible number of states and transitions. These algorithms generalize classical automata determinization and minimization to deal properly with the probabilities of alternative hypotheses and with the relationships between units (distributions, phones, words) at different levels in the recognition system. We illustrate their use in several applications, and report the results of our experiments.",
author = "Mehryar Mohri and Michael Riley",
year = "1999",
month = "5",
doi = "10.1016/S0167-6393(98)00043-0",
language = "English (US)",
volume = "28",
pages = "1--12",
journal = "Speech Communication",
issn = "0167-6393",
publisher = "Elsevier",
number = "1",

}

TY - JOUR

T1 - Network optimizations for large-vocabulary speech recognition

AU - Mohri, Mehryar

AU - Riley, Michael

PY - 1999/5

Y1 - 1999/5

N2 - The redundancy and the size of networks in large-vocabulary speech recognition systems can have a critical effect on their overall performance. We describe the use of two new algorithms: weighted determinization and minimization. These algorithms transform recognition labeled networks into equivalent ones that require much less time and space in large-vocabulary speech recognition. They are both optimal: weighted determinization eliminates the number of alternatives at each state to the minimum, and weighted minimization reduces the size of deterministic networks to the smallest possible number of states and transitions. These algorithms generalize classical automata determinization and minimization to deal properly with the probabilities of alternative hypotheses and with the relationships between units (distributions, phones, words) at different levels in the recognition system. We illustrate their use in several applications, and report the results of our experiments.

AB - The redundancy and the size of networks in large-vocabulary speech recognition systems can have a critical effect on their overall performance. We describe the use of two new algorithms: weighted determinization and minimization. These algorithms transform recognition labeled networks into equivalent ones that require much less time and space in large-vocabulary speech recognition. They are both optimal: weighted determinization eliminates the number of alternatives at each state to the minimum, and weighted minimization reduces the size of deterministic networks to the smallest possible number of states and transitions. These algorithms generalize classical automata determinization and minimization to deal properly with the probabilities of alternative hypotheses and with the relationships between units (distributions, phones, words) at different levels in the recognition system. We illustrate their use in several applications, and report the results of our experiments.

UR - http://www.scopus.com/inward/record.url?scp=0032661656&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0032661656&partnerID=8YFLogxK

U2 - 10.1016/S0167-6393(98)00043-0

DO - 10.1016/S0167-6393(98)00043-0

M3 - Article

VL - 28

SP - 1

EP - 12

JO - Speech Communication

JF - Speech Communication

SN - 0167-6393

IS - 1

ER -