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 montre comment un algorithme de regroupement d'états de division qui génère des modèles acoustiques de Markov cachés (HMM) peut bénéficier d'une représentation à mélange lié de la fonction de densité de probabilité (pdf) d'un état et augmenter les performances de reconnaissance. Les algorithmes de clustering populaires basés sur des arbres de décision, comme par exemple l'algorithme de division d'état successif (SSS), utilisent une simplification lors du clustering des données. Ils représentent un état utilisant un seul pdf gaussien. Nous montrons que cette approximation de la vraie pdf par une seule gaussienne est trop grossière, par exemple une seule gaussienne ne peut pas représenter les différences dans les parties symétriques des pdf des nouveaux états hypothétiques générés lors de l'évaluation du gain de partage d'état (qui déterminera le division de l'État). L’utilisation de représentations plus sophistiquées conduirait à des problèmes de calcul insolubles que nous résolvons en utilisant une représentation pdf à mélange lié. De plus, nous contraignons le livre de codes à être immuable pendant la scission. Entre les divisions d'état, cette contrainte est assouplie et le livre de codes est mis à jour. Dans cet article, nous proposons donc une extension à l’algorithme SSS, appelé algorithme Tied-mixture Successive State Splitting (TM-SSS). TM-SSS affiche une réduction d'erreur jusqu'à environ 31 % par rapport à l'algorithme ML-SSS (Maximum-Likelihood Successive State Split) pour une expérience de reconnaissance de mots.
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
Alexandre GIRARDI, Harald SINGER, Kiyohiro SHIKANO, Satoshi NAKAMURA, "Maximum Likelihood Successive State Splitting Algorithm for Tied-Mixture HMnet" in IEICE TRANSACTIONS on Information,
vol. E83-D, no. 10, pp. 1890-1897, October 2000, doi: .
Abstract: This paper shows how a divisive state clustering algorithm that generates acoustic Hidden Markov models (HMM) can benefit from a tied-mixture representation of the probability density function (pdf) of a state and increase the recognition performance. Popular decision tree based clustering algorithms, like for example the Successive State Splitting algorithm (SSS) make use of a simplification when clustering data. They represent a state using a single Gaussian pdf. We show that this approximation of the true pdf by a single Gaussian is too coarse, for example a single Gaussian cannot represent the differences in the symmetric parts of the pdf's of the new hypothetical states generated when evaluating the state split gain (which will determine the state split). The use of more sophisticated representations would lead to intractable computational problems that we solve by using a tied-mixture pdf representation. Additionally, we constrain the codebook to be immutable during the split. Between state splits, this constraint is relaxed and the codebook is updated. In this paper, we thus propose an extension to the SSS algorithm, the so-called Tied-mixture Successive State Splitting algorithm (TM-SSS). TM-SSS shows up to about 31% error reduction in comparison with Maximum-Likelihood Successive State Split algorithm (ML-SSS) for a word recognition experiment.
URL: https://global.ieice.org/en_transactions/information/10.1587/e83-d_10_1890/_p
Copier
@ARTICLE{e83-d_10_1890,
author={Alexandre GIRARDI, Harald SINGER, Kiyohiro SHIKANO, Satoshi NAKAMURA, },
journal={IEICE TRANSACTIONS on Information},
title={Maximum Likelihood Successive State Splitting Algorithm for Tied-Mixture HMnet},
year={2000},
volume={E83-D},
number={10},
pages={1890-1897},
abstract={This paper shows how a divisive state clustering algorithm that generates acoustic Hidden Markov models (HMM) can benefit from a tied-mixture representation of the probability density function (pdf) of a state and increase the recognition performance. Popular decision tree based clustering algorithms, like for example the Successive State Splitting algorithm (SSS) make use of a simplification when clustering data. They represent a state using a single Gaussian pdf. We show that this approximation of the true pdf by a single Gaussian is too coarse, for example a single Gaussian cannot represent the differences in the symmetric parts of the pdf's of the new hypothetical states generated when evaluating the state split gain (which will determine the state split). The use of more sophisticated representations would lead to intractable computational problems that we solve by using a tied-mixture pdf representation. Additionally, we constrain the codebook to be immutable during the split. Between state splits, this constraint is relaxed and the codebook is updated. In this paper, we thus propose an extension to the SSS algorithm, the so-called Tied-mixture Successive State Splitting algorithm (TM-SSS). TM-SSS shows up to about 31% error reduction in comparison with Maximum-Likelihood Successive State Split algorithm (ML-SSS) for a word recognition experiment.},
keywords={},
doi={},
ISSN={},
month={October},}
Copier
TY - JOUR
TI - Maximum Likelihood Successive State Splitting Algorithm for Tied-Mixture HMnet
T2 - IEICE TRANSACTIONS on Information
SP - 1890
EP - 1897
AU - Alexandre GIRARDI
AU - Harald SINGER
AU - Kiyohiro SHIKANO
AU - Satoshi NAKAMURA
PY - 2000
DO -
JO - IEICE TRANSACTIONS on Information
SN -
VL - E83-D
IS - 10
JA - IEICE TRANSACTIONS on Information
Y1 - October 2000
AB - This paper shows how a divisive state clustering algorithm that generates acoustic Hidden Markov models (HMM) can benefit from a tied-mixture representation of the probability density function (pdf) of a state and increase the recognition performance. Popular decision tree based clustering algorithms, like for example the Successive State Splitting algorithm (SSS) make use of a simplification when clustering data. They represent a state using a single Gaussian pdf. We show that this approximation of the true pdf by a single Gaussian is too coarse, for example a single Gaussian cannot represent the differences in the symmetric parts of the pdf's of the new hypothetical states generated when evaluating the state split gain (which will determine the state split). The use of more sophisticated representations would lead to intractable computational problems that we solve by using a tied-mixture pdf representation. Additionally, we constrain the codebook to be immutable during the split. Between state splits, this constraint is relaxed and the codebook is updated. In this paper, we thus propose an extension to the SSS algorithm, the so-called Tied-mixture Successive State Splitting algorithm (TM-SSS). TM-SSS shows up to about 31% error reduction in comparison with Maximum-Likelihood Successive State Split algorithm (ML-SSS) for a word recognition experiment.
ER -