Presentation	2000/1/21 Automatic Creation of Mixture Structure by Successive State Split Masaru Takano,
PDF Download Page	PDF download Page Link
Abstract(in Japanese)	(See Japanese page)
Abstract(in English)	In most cases of making an acoustic model, at least two parameters are needed, the number of states and the number of mixture components, and both are difficult to decide properly. In this report, we propose a method for deciding them simultaneously with only one parameter, the total number of Gaussians, while the HMM topology is made via a multi-path HMM. This method consists of two steps:making the multi-path HMM and converting it into a single-path HMM.The first step is a variant of ML-SSS. This step makes a multi-path HMM by iterating a contextual split and a path split. Each state in the HMM is regrouped in the second step to make a single-path HMM. In a recognition experiment using Japanese spontaneous speech, the models made by the proposed method achieved better performance than a conventional model and the proposed used less than half as many Gaussians.
Keyword(in Japanese)	(See Japanese page)
Keyword(in English)	acoustic model / HMM / mixture distribution / SSS / topology
Paper #	SP99-141
Date of Issue

Conference Information
Committee	SP
Conference Date	2000/1/21(1days)
Place (in Japanese)	(See Japanese page)
Place (in English)
Topics (in Japanese)	(See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To	Speech (SP)
Language	JPN
Title (in Japanese)	(See Japanese page)
Sub Title (in Japanese)	(See Japanese page)
Title (in English)	Automatic Creation of Mixture Structure by Successive State Split
Sub Title (in English)
Keyword(1)	acoustic model
Keyword(2)	HMM
Keyword(3)	mixture distribution
Keyword(4)	SSS
Keyword(5)	topology
1st Author's Name	Masaru Takano
1st Author's Affiliation	ATR Interpreting Telecommunications Research Laboratories()
Date	2000/1/21
Paper #	SP99-141
Volume (vol)	vol.99
Number (no)	577
Page	pp.pp.-
#Pages	6
Date of Issue