LARGE VOCABULARY SPEAKER-INDEPENDENT CONTINUOUS SPEECH RECOGNITION WITH A NEW HYBRID SYSTEM BASED ON MMI-NEURAL NETWORKS

G. Rigoll; Ch Neukirchen; J. Rottland

LARGE VOCABULARY SPEAKER-INDEPENDENT CONTINUOUS SPEECH RECOGNITION WITH A NEW HYBRID SYSTEM BASED ON MMI-NEURAL NETWORKS

G. Rigoll, Ch Neukirchen, J. Rottland

University of Duisburg-Essen

Research output: Contribution to conference › Paper › peer-review

3 Scopus citations

Abstract

This paper presents a new hybrid system for speaker independent continuous speech recognition in a large vocabulary task. The hybrid system is a combination of context dependent discrete Hidden Markov Models and artificial neural networks that are trained by an information theory based algorithm. This algorithm maximizes the Mutual /nformation (MMI) between the network output and the phone descriptions by applying a self-organizing learning approach instead of forcing constrained network outputs. Recognition results have shown that the new hybrid system outperforms a classical k-means-VQ-based HMM-system. For the speaker independent DARPA Resource Management (RM) task (perplexity 60) we report a decrease in word recognition error rate up to 35% (close to the best continuous pdf systems).

Original language	English
Pages	1659-1662
Number of pages	4
State	Published - 1995
Externally published	Yes
Event	4th European Conference on Speech Communication and Technology, EUROSPEECH 1995 - Madrid, Spain Duration: 18 Sep 1995 → 21 Sep 1995

Conference

Conference	4th European Conference on Speech Communication and Technology, EUROSPEECH 1995
Country/Territory	Spain
City	Madrid
Period	18/09/95 → 21/09/95

Cite this

@conference{ea01e61d9fdf4896be882891cd949c85,

title = "LARGE VOCABULARY SPEAKER-INDEPENDENT CONTINUOUS SPEECH RECOGNITION WITH A NEW HYBRID SYSTEM BASED ON MMI-NEURAL NETWORKS",

abstract = "This paper presents a new hybrid system for speaker independent continuous speech recognition in a large vocabulary task. The hybrid system is a combination of context dependent discrete Hidden Markov Models and artificial neural networks that are trained by an information theory based algorithm. This algorithm maximizes the Mutual /nformation (MMI) between the network output and the phone descriptions by applying a self-organizing learning approach instead of forcing constrained network outputs. Recognition results have shown that the new hybrid system outperforms a classical k-means-VQ-based HMM-system. For the speaker independent DARPA Resource Management (RM) task (perplexity 60) we report a decrease in word recognition error rate up to 35% (close to the best continuous pdf systems).",

author = "G. Rigoll and Ch Neukirchen and J. Rottland",

note = "Publisher Copyright: {\textcopyright} 1995 4th European Conference on Speech Communication and Technology, EUROSPEECH 1995. All rights reserved.; 4th European Conference on Speech Communication and Technology, EUROSPEECH 1995 ; Conference date: 18-09-1995 Through 21-09-1995",

year = "1995",

language = "English",

pages = "1659--1662",

}

TY - CONF

T1 - LARGE VOCABULARY SPEAKER-INDEPENDENT CONTINUOUS SPEECH RECOGNITION WITH A NEW HYBRID SYSTEM BASED ON MMI-NEURAL NETWORKS

AU - Rigoll, G.

AU - Neukirchen, Ch

AU - Rottland, J.

PY - 1995

Y1 - 1995

N2 - This paper presents a new hybrid system for speaker independent continuous speech recognition in a large vocabulary task. The hybrid system is a combination of context dependent discrete Hidden Markov Models and artificial neural networks that are trained by an information theory based algorithm. This algorithm maximizes the Mutual /nformation (MMI) between the network output and the phone descriptions by applying a self-organizing learning approach instead of forcing constrained network outputs. Recognition results have shown that the new hybrid system outperforms a classical k-means-VQ-based HMM-system. For the speaker independent DARPA Resource Management (RM) task (perplexity 60) we report a decrease in word recognition error rate up to 35% (close to the best continuous pdf systems).

AB - This paper presents a new hybrid system for speaker independent continuous speech recognition in a large vocabulary task. The hybrid system is a combination of context dependent discrete Hidden Markov Models and artificial neural networks that are trained by an information theory based algorithm. This algorithm maximizes the Mutual /nformation (MMI) between the network output and the phone descriptions by applying a self-organizing learning approach instead of forcing constrained network outputs. Recognition results have shown that the new hybrid system outperforms a classical k-means-VQ-based HMM-system. For the speaker independent DARPA Resource Management (RM) task (perplexity 60) we report a decrease in word recognition error rate up to 35% (close to the best continuous pdf systems).

UR - http://www.scopus.com/inward/record.url?scp=0001789995&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:0001789995

SP - 1659

EP - 1662

T2 - 4th European Conference on Speech Communication and Technology, EUROSPEECH 1995

Y2 - 18 September 1995 through 21 September 1995

ER -

LARGE VOCABULARY SPEAKER-INDEPENDENT CONTINUOUS SPEECH RECOGNITION WITH A NEW HYBRID SYSTEM BASED ON MMI-NEURAL NETWORKS

Abstract

Conference

Other files and links

Fingerprint

Cite this