Model parameter estimation for mixture density polynomial segment models

T. Fukada, K. K. Paliwal, Yoshinori Sagisaka

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

In this paper, we propose parameter estimation techniques for mixture density polynomial segment models (MDPSMs) where their trajectories are specified with an arbitrary regression order. MDPSM parameters can be trained in one of three different ways: (1) segment clustering; (2) expectation maximization (EM) training of mean trajectories; and (3) EM training of mean and variance trajectories. These parameter estimation methods were evaluated in TIMIT vowel classification experiments. The experimental results showed that modelling both the mean and variance trajectories is consistently superior to modelling only the mean trajectory. We also found that modelling both trajectories results in significant improvements over the conventional HMM.

Original languageEnglish
Pages (from-to)229-246
Number of pages18
JournalComputer Speech and Language
Volume12
Issue number3
Publication statusPublished - 1998 Jun
Externally publishedYes

Fingerprint

Statistical Models
Parameter estimation
Parameter Estimation
Trajectories
Trajectory
Polynomial
Cluster Analysis
Expectation Maximization
estimation procedure
Modeling
Model
regression
experiment
Regression
Clustering
Experimental Results
Arbitrary
Experiment
Experiments

ASJC Scopus subject areas

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

Cite this

Model parameter estimation for mixture density polynomial segment models. / Fukada, T.; Paliwal, K. K.; Sagisaka, Yoshinori.

In: Computer Speech and Language, Vol. 12, No. 3, 06.1998, p. 229-246.

Research output: Contribution to journalArticle

@article{1d136997ab4d4a448ffb60785d09d299,
title = "Model parameter estimation for mixture density polynomial segment models",
abstract = "In this paper, we propose parameter estimation techniques for mixture density polynomial segment models (MDPSMs) where their trajectories are specified with an arbitrary regression order. MDPSM parameters can be trained in one of three different ways: (1) segment clustering; (2) expectation maximization (EM) training of mean trajectories; and (3) EM training of mean and variance trajectories. These parameter estimation methods were evaluated in TIMIT vowel classification experiments. The experimental results showed that modelling both the mean and variance trajectories is consistently superior to modelling only the mean trajectory. We also found that modelling both trajectories results in significant improvements over the conventional HMM.",
author = "T. Fukada and Paliwal, {K. K.} and Yoshinori Sagisaka",
year = "1998",
month = "6",
language = "English",
volume = "12",
pages = "229--246",
journal = "Computer Speech and Language",
issn = "0885-2308",
publisher = "Academic Press Inc.",
number = "3",

}

TY - JOUR

T1 - Model parameter estimation for mixture density polynomial segment models

AU - Fukada, T.

AU - Paliwal, K. K.

AU - Sagisaka, Yoshinori

PY - 1998/6

Y1 - 1998/6

N2 - In this paper, we propose parameter estimation techniques for mixture density polynomial segment models (MDPSMs) where their trajectories are specified with an arbitrary regression order. MDPSM parameters can be trained in one of three different ways: (1) segment clustering; (2) expectation maximization (EM) training of mean trajectories; and (3) EM training of mean and variance trajectories. These parameter estimation methods were evaluated in TIMIT vowel classification experiments. The experimental results showed that modelling both the mean and variance trajectories is consistently superior to modelling only the mean trajectory. We also found that modelling both trajectories results in significant improvements over the conventional HMM.

AB - In this paper, we propose parameter estimation techniques for mixture density polynomial segment models (MDPSMs) where their trajectories are specified with an arbitrary regression order. MDPSM parameters can be trained in one of three different ways: (1) segment clustering; (2) expectation maximization (EM) training of mean trajectories; and (3) EM training of mean and variance trajectories. These parameter estimation methods were evaluated in TIMIT vowel classification experiments. The experimental results showed that modelling both the mean and variance trajectories is consistently superior to modelling only the mean trajectory. We also found that modelling both trajectories results in significant improvements over the conventional HMM.

UR - http://www.scopus.com/inward/record.url?scp=0042856578&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0042856578&partnerID=8YFLogxK

M3 - Article

VL - 12

SP - 229

EP - 246

JO - Computer Speech and Language

JF - Computer Speech and Language

SN - 0885-2308

IS - 3

ER -