Multi-Condition Training for Unknown Environment Adaptation in Robust ASR Under Real Conditions

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F09%3A00157917" target="_blank" >RIV/68407700:21230/09:00157917 - isvavai.cz</a>

Result on the web

—

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Multi-Condition Training for Unknown Environment Adaptation in Robust ASR Under Real Conditions

Original language description

Automatic speech recognition (ASR) systems frequently work in noisy environment. As they are often trained on clean speech data, noise reduction or adaptation techniques are applied to decrease the influence of background disturbance. This paper analysesthe recognition performance within such adaptation when multi-condition training data from real environment are used for training initial models. Although the quality of such models can decrease with the presence of noise in the training material, theyare supposed to include initial information about noise and consequently support the adaptation procedure. Experimental results show significant improvement of the proposed training method in robust ASR task under unknown noisy conditions. The decrease by 29% and 14% in word error rate against the case with clean speech training data was reached for the non-adapted and adapted system respectively.

Czech name

—

Czech description

—

Type

J_x - Unclassified - Peer-reviewed scientific article (Jimp, Jsc and Jost)

CEP classification

JA - Electronics and optoelectronics

OECD FORD branch

—

Project

Result was created during the realization of more than one project. More information in the Projects tab.

Continuities

Z - Vyzkumny zamer (s odkazem do CEZ)

Publication year

2009

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Name of the periodical

Acta Polytechnica

ISSN

1210-2709

e-ISSN

—

Volume of the periodical

49

Issue of the periodical within the volume

2-3/2009

Country of publishing house

CZ - CZECH REPUBLIC

Number of pages

5

Pages from-to

—

UT code for WoS article

—

EID of the result in the Scopus database

—