Speaker adaptation in the Czech large vocabulary continuous speech recognition system

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F49777513%3A23520%2F01%3A00064720" target="_blank" >RIV/49777513:23520/01:00064720 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Speaker adaptation in the Czech large vocabulary continuous speech recognition system

Popis výsledku v původním jazyce

This paper presents a method for improvement of a recognition rate in the Czech large vocabulary continuous speech recognition system. Speaker independent systems are usually trained by the speech data of many speakers but achieve a lower recognition rate than speaker dependent systems. A need for a large amount of the speech training data from each speaker makes the latter practically unusable. Speaker adaptation is a general name for a group of methods which try to increase a recognition rate of the SI system using only a small amount of training speech data from a new speaker. The maximum a posteriori method (based on Bayesian approach) together with its necessary corrections for the Czech language (the parameter tying method) is presented. Resultsof experiments that were carried out to verify performance of presented methods are discussed in the last part of this article.

Název v anglickém jazyce

Speaker adaptation in the Czech large vocabulary continuous speech recognition system

Popis výsledku anglicky

This paper presents a method for improvement of a recognition rate in the Czech large vocabulary continuous speech recognition system. Speaker independent systems are usually trained by the speech data of many speakers but achieve a lower recognition rate than speaker dependent systems. A need for a large amount of the speech training data from each speaker makes the latter practically unusable. Speaker adaptation is a general name for a group of methods which try to increase a recognition rate of the SI system using only a small amount of training speech data from a new speaker. The maximum a posteriori method (based on Bayesian approach) together with its necessary corrections for the Czech language (the parameter tying method) is presented. Resultsof experiments that were carried out to verify performance of presented methods are discussed in the last part of this article.

Druh

D - Stať ve sborníku

CEP obor

JD - Využití počítačů, robotika a její aplikace

OECD FORD obor

—

Projekt

<a href="/cs/project/GV102%2F96%2FK087" target="_blank" >GV102/96/K087: Teorie a aplikace hlasové komunikace v Češtině</a><br>

Návaznosti

Z - Vyzkumny zamer (s odkazem do CEZ)

Rok uplatnění

2001

Kód důvěrnosti údajů

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

Název statě ve sborníku

Speaker adaptation in the Czech large vocabulary continuous speech recognition system

ISBN

—

ISSN

—

e-ISSN

—

Počet stran výsledku

3

Strana od-do

—

Název nakladatele

Université Paul Sabatier

Místo vydání

Toulouse

Místo konání akce

Toulouse

Datum konání akce

1. 1. 2001

Typ akce podle státní příslušnosti

CST - Celostátní akce

Kód UT WoS článku

—