Multilingual acoustic modeling for speech recognition based on Subspace Gaussian Mixture Models

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26230%2F10%3APU91981" target="_blank" >RIV/00216305:26230/10:PU91981 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Multilingual acoustic modeling for speech recognition based on Subspace Gaussian Mixture Models

Popis výsledku v původním jazyce

Although research has previously been done on multilingual speech recognition, it has been found to be very difficult to improve over separately trained systems. The usual approach has been to use some kind of &quot;universal phone set&quot; that coversmultiple languages. We report experiments on a different approach to multilingual speech recognition, in which the phone sets are entirely distinct but the model has parameters not tied to specific states that are shared across languages. We use a modelcalled a &quot;Subspace Gaussian Mixture Model&quot; where states' distributions are Gaussian Mixture Models with a common structure, constrained to lie in a subspace of the total parameter space. The parameters that define this subspace can be shared across languages. We obtain substantial WER improvements with this approach, especially with very small amounts of inlanguage training data.

Název v anglickém jazyce

Multilingual acoustic modeling for speech recognition based on Subspace Gaussian Mixture Models

Popis výsledku anglicky

Although research has previously been done on multilingual speech recognition, it has been found to be very difficult to improve over separately trained systems. The usual approach has been to use some kind of &quot;universal phone set&quot; that coversmultiple languages. We report experiments on a different approach to multilingual speech recognition, in which the phone sets are entirely distinct but the model has parameters not tied to specific states that are shared across languages. We use a modelcalled a &quot;Subspace Gaussian Mixture Model&quot; where states' distributions are Gaussian Mixture Models with a common structure, constrained to lie in a subspace of the total parameter space. The parameters that define this subspace can be shared across languages. We obtain substantial WER improvements with this approach, especially with very small amounts of inlanguage training data.

Druh

D - Stať ve sborníku

CEP obor

IN - Informatika

OECD FORD obor

—

Projekt

<a href="/cs/project/FR-TI1%2F034" target="_blank" >FR-TI1/034: Multilingvální rozpoznávání a vyhledávání v řeči pro elektronické slovníky</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2010

Kód důvěrnosti údajů

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

Název statě ve sborníku

Proc. International Conference on Acoustictics, Speech, and Signal Processing

ISBN

978-1-4244-4296-6

ISSN

—

e-ISSN

—

Počet stran výsledku

4

Strana od-do

—

Název nakladatele

IEEE Signal Processing Society

Místo vydání

Dallas

Místo konání akce

Dallas

Datum konání akce

14. 3. 2010

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—