Skip to content

Publications

Detail of publication

Citation

Zbyněk Zajíc and Machlica Lukáš and Müller Luděk : Initialization of Adaptation by Sufficient Statistics Using Phonetic Tree . IEEE 11th International Conference on Signal Processing, p. 503-506, Beijing, 2012.

Download PDF

PDF

Abstract

In this work we deal with the problem of small amount of data when estimating a feature transformation for the speaker adaptation of an acoustic model. Our goal is to compensate for the lack of adaptation data by a proper initialization of transformation matrices. Methods used in such situations are described, they are based on collecting additional accumulated statistics from nearest speakers. The proposed initialization approach is based on accumulated statistics too, but it incorporates also phonetic information when selecting the “nearest” statistics. Initialization methods compensating for the absence of actual speaker’s data are tested on telephone recordings with different amounts of adaptation data. In worst situation with extremely small amount of adaptation data relative improvement of 5% is obtained.

Detail of publication

Title: Initialization of Adaptation by Sufficient Statistics Using Phonetic Tree
Author: Zbyněk Zajíc ; Machlica Lukáš ; Müller Luděk
Language: English
Year: 2012
Type of publication: Papers in proceedings of reviewed conferences
Title of journal or book: IEEE 11th International Conference on Signal Processing
Page: 503 - 506
Address: Beijing
/ 2013-02-18 16:34:32 /

Keywords

speech recognition, adaptation, initialization, phonetic tree

BibTeX

@MISC{ZbynekZajic_2012_Initializationof,
 author = {Zbyn\v{e}k Zaj\'{i}c and Machlica Luk\'{a}\v{s} and M\"{u}ller Lud\v{e}k},
 title = {Initialization of Adaptation by Sufficient Statistics Using Phonetic Tree},
 year = {2012},
 journal = {IEEE 11th International Conference on Signal Processing},
 address = {Beijing},
 pages = {503-506},
 url = {http://www.kky.zcu.cz/en/publications/ZbynekZajic_2012_Initializationof},
}