Enhancing phoneme recogniser performance with a simple rule-based language model

Pertti Väyrynen; Johannes Peltola; Tapio Seppänen

Enhancing phoneme recogniser performance with a simple rule-based language model

Pertti Väyrynen, Johannes Peltola, Tapio Seppänen

University of Oulu

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

Abstract

The phoneme classification inaccuracy at the acoustic phonetic level is a major weakness in most speech recognition systems. However, the inaccuracy will violate phonotactic constraints at the acoustic phonetic level. A better performance is expected if a language model is adopted in a recognition system for post-processing phoneme estimates and making corrections with a set of explicit rules of the language used. We developed a simple language model with a set of rules to correct phoneme classification errors made by a Hidden Markov Model based phoneme recognizer. The experimental results indicate that about 20% of
recognition errors can be corrected.

Original language	English
Title of host publication	STeP2000 Millennium of Artificial Intelligence
Subtitle of host publication	Al of Tommorrow
Pages	171-178
Number of pages	7
Publication status	Published - 2000
MoE publication type	A4 Article in a conference publication
Event	9th Finnish Artificial Intelligence Conference : Millennium of Artificial Intelligence - Espoo, Finland Duration: 28 Aug 2000 → 30 Aug 2000 Conference number: 9

Conference

Conference	9th Finnish Artificial Intelligence Conference
Abbreviated title	STeP 2000
Country/Territory	Finland
City	Espoo
Period	28/08/00 → 30/08/00

Cite this

@inproceedings{3deec90e27174cc19202f92c243d9c5f,

title = "Enhancing phoneme recogniser performance with a simple rule-based language model",

abstract = "The phoneme classification inaccuracy at the acoustic phonetic level is a major weakness in most speech recognition systems. However, the inaccuracy will violate phonotactic constraints at the acoustic phonetic level. A better performance is expected if a language model is adopted in a recognition system for post-processing phoneme estimates and making corrections with a set of explicit rules of the language used. We developed a simple language model with a set of rules to correct phoneme classification errors made by a Hidden Markov Model based phoneme recognizer. The experimental results indicate that about 20% ofrecognition errors can be corrected.",

author = "Pertti V{\"a}yrynen and Johannes Peltola and Tapio Sepp{\"a}nen",

year = "2000",

language = "English",

isbn = "951-22-5128-0",

pages = "171--178",

booktitle = "STeP2000 Millennium of Artificial Intelligence",

note = "9th Finnish Artificial Intelligence Conference : Millennium of Artificial Intelligence, STeP 2000 ; Conference date: 28-08-2000 Through 30-08-2000",

}

TY - GEN

T1 - Enhancing phoneme recogniser performance with a simple rule-based language model

AU - Väyrynen, Pertti

AU - Peltola, Johannes

AU - Seppänen, Tapio

N1 - Conference code: 9

PY - 2000

Y1 - 2000

N2 - The phoneme classification inaccuracy at the acoustic phonetic level is a major weakness in most speech recognition systems. However, the inaccuracy will violate phonotactic constraints at the acoustic phonetic level. A better performance is expected if a language model is adopted in a recognition system for post-processing phoneme estimates and making corrections with a set of explicit rules of the language used. We developed a simple language model with a set of rules to correct phoneme classification errors made by a Hidden Markov Model based phoneme recognizer. The experimental results indicate that about 20% ofrecognition errors can be corrected.

AB - The phoneme classification inaccuracy at the acoustic phonetic level is a major weakness in most speech recognition systems. However, the inaccuracy will violate phonotactic constraints at the acoustic phonetic level. A better performance is expected if a language model is adopted in a recognition system for post-processing phoneme estimates and making corrections with a set of explicit rules of the language used. We developed a simple language model with a set of rules to correct phoneme classification errors made by a Hidden Markov Model based phoneme recognizer. The experimental results indicate that about 20% ofrecognition errors can be corrected.

M3 - Conference article in proceedings

SN - 951-22-5128-0

SP - 171

EP - 178

BT - STeP2000 Millennium of Artificial Intelligence

T2 - 9th Finnish Artificial Intelligence Conference

Y2 - 28 August 2000 through 30 August 2000

ER -

Enhancing phoneme recogniser performance with a simple rule-based language model

Abstract

Conference

Fingerprint

Cite this