The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Cet article propose une combinaison efficace de recyclage de vraisemblance d'état et de calcul de vraisemblance d'état par lots pour accélérer le calcul de vraisemblance acoustique dans un système de reconnaissance vocale basé sur HMM. Le recyclage et le calcul des lots reposent chacun sur des approches techniques différentes : le premier est une technique purement algorithmique tandis que le second exploite pleinement l'architecture informatique. Pour accélérer davantage le processus de reconnaissance en les combinant efficacement, nous introduisons traitement rapide conditionnel et à la recul acoustique. Le traitement rapide conditionnel repose sur deux critères. La première activité potentielle Ce critère est utilisé pour contrôler non seulement le recyclage des vraisemblances d'état dans la trame actuelle, mais également le précalcul des vraisemblances d'état pour plusieurs trames successives. La deuxième fiabilité Le critère et l'atténuation acoustique sont utilisés pour contrôler le choix des probabilités d'état recyclées ou calculées par lots lorsqu'elles sont contradictoires dans la combinaison et pour empêcher la précision des mots de se dégrader. Des expériences de reconnaissance vocale spontanée à grand vocabulaire utilisant quatre machines à processeur différentes dans deux conditions environnementales ont montré que, par rapport au système de reconnaissance de base, au recyclage et au calcul par lots, notre technique d'accélération combinée réduisait encore davantage le temps de calcul de la vraisemblance acoustique et le temps total de reconnaissance. Nous avons également effectué des analyses détaillées pour révéler les mécanismes d'accélération et de dépendance environnementale de chaque technique en classant les types de probabilités d'état et en comptant chacun d'eux. Les résultats de l'analyse ont confirmé l'efficacité de la technique d'accélération combinée.
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Atsunori OGAWA, Satoshi TAKAHASHI, Atsushi NAKAMURA, "Efficient Combination of Likelihood Recycling and Batch Calculation for Fast Acoustic Likelihood Calculation" in IEICE TRANSACTIONS on Information,
vol. E94-D, no. 3, pp. 648-658, March 2011, doi: 10.1587/transinf.E94.D.648.
Abstract: This paper proposes an efficient combination of state likelihood recycling and batch state likelihood calculation for accelerating acoustic likelihood calculation in an HMM-based speech recognizer. Recycling and batch calculation are each based on different technical approaches, i.e. the former is a purely algorithmic technique while the latter fully exploits computer architecture. To accelerate the recognition process further by combining them efficiently, we introduce conditional fast processing and acoustic backing-off. Conditional fast processing is based on two criteria. The first potential activity criterion is used to control not only the recycling of state likelihoods at the current frame but also the precalculation of state likelihoods for several succeeding frames. The second reliability criterion and acoustic backing-off are used to control the choice of recycled or batch calculated state likelihoods when they are contradictory in the combination and to prevent word accuracies from degrading. Large vocabulary spontaneous speech recognition experiments using four different CPU machines under two environmental conditions showed that, compared with the baseline recognizer, recycling and batch calculation, our combined acceleration technique further reduced both of the acoustic likelihood calculation time and the total recognition time. We also performed detailed analyses to reveal each technique's acceleration and environmental dependency mechanisms by classifying types of state likelihoods and counting each of them. The analysis results comfirmed the effectiveness of the combined acceleration technique.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.E94.D.648/_p
Copier
@ARTICLE{e94-d_3_648,
author={Atsunori OGAWA, Satoshi TAKAHASHI, Atsushi NAKAMURA, },
journal={IEICE TRANSACTIONS on Information},
title={Efficient Combination of Likelihood Recycling and Batch Calculation for Fast Acoustic Likelihood Calculation},
year={2011},
volume={E94-D},
number={3},
pages={648-658},
abstract={This paper proposes an efficient combination of state likelihood recycling and batch state likelihood calculation for accelerating acoustic likelihood calculation in an HMM-based speech recognizer. Recycling and batch calculation are each based on different technical approaches, i.e. the former is a purely algorithmic technique while the latter fully exploits computer architecture. To accelerate the recognition process further by combining them efficiently, we introduce conditional fast processing and acoustic backing-off. Conditional fast processing is based on two criteria. The first potential activity criterion is used to control not only the recycling of state likelihoods at the current frame but also the precalculation of state likelihoods for several succeeding frames. The second reliability criterion and acoustic backing-off are used to control the choice of recycled or batch calculated state likelihoods when they are contradictory in the combination and to prevent word accuracies from degrading. Large vocabulary spontaneous speech recognition experiments using four different CPU machines under two environmental conditions showed that, compared with the baseline recognizer, recycling and batch calculation, our combined acceleration technique further reduced both of the acoustic likelihood calculation time and the total recognition time. We also performed detailed analyses to reveal each technique's acceleration and environmental dependency mechanisms by classifying types of state likelihoods and counting each of them. The analysis results comfirmed the effectiveness of the combined acceleration technique.},
keywords={},
doi={10.1587/transinf.E94.D.648},
ISSN={1745-1361},
month={March},}
Copier
TY - JOUR
TI - Efficient Combination of Likelihood Recycling and Batch Calculation for Fast Acoustic Likelihood Calculation
T2 - IEICE TRANSACTIONS on Information
SP - 648
EP - 658
AU - Atsunori OGAWA
AU - Satoshi TAKAHASHI
AU - Atsushi NAKAMURA
PY - 2011
DO - 10.1587/transinf.E94.D.648
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E94-D
IS - 3
JA - IEICE TRANSACTIONS on Information
Y1 - March 2011
AB - This paper proposes an efficient combination of state likelihood recycling and batch state likelihood calculation for accelerating acoustic likelihood calculation in an HMM-based speech recognizer. Recycling and batch calculation are each based on different technical approaches, i.e. the former is a purely algorithmic technique while the latter fully exploits computer architecture. To accelerate the recognition process further by combining them efficiently, we introduce conditional fast processing and acoustic backing-off. Conditional fast processing is based on two criteria. The first potential activity criterion is used to control not only the recycling of state likelihoods at the current frame but also the precalculation of state likelihoods for several succeeding frames. The second reliability criterion and acoustic backing-off are used to control the choice of recycled or batch calculated state likelihoods when they are contradictory in the combination and to prevent word accuracies from degrading. Large vocabulary spontaneous speech recognition experiments using four different CPU machines under two environmental conditions showed that, compared with the baseline recognizer, recycling and batch calculation, our combined acceleration technique further reduced both of the acoustic likelihood calculation time and the total recognition time. We also performed detailed analyses to reveal each technique's acceleration and environmental dependency mechanisms by classifying types of state likelihoods and counting each of them. The analysis results comfirmed the effectiveness of the combined acceleration technique.
ER -