Automatická segmentace mluvené řeči do větných jednotek

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F49777513%3A23520%2F08%3A00500610" target="_blank" >RIV/49777513:23520/08:00500610 - isvavai.cz</a>

Výsledek na webu

—

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Automatic Segmentation of Speech into Sentence-like Units

Popis výsledku v původním jazyce

This thesis deals with the problem of automatic segmentation of speech recognition output into sentence-like units. The work is focused on two languages - English and Czech. First, I describe creation of two Czech speech corpora with structural metadataannotation in two different domains: broadcast news and broadcast conversations. The main goal of this work is to develop automatic systems for dialog act segmentation of English multiparty meetings and sentence unit segmentation of the two new Czech corpora. I use and compare three modeling approaches - hidden Markov models, maximum entropy, and a boosting-based algorithm called BoosTexter. All of these approaches rely on two information sources - recognized words and prosody. In addition, I explore speaker adaptation for this task. The results indicate that superior performance is achieved when the three statistical models are combined via posterior probability interpolation.

Název v anglickém jazyce

Automatic Segmentation of Speech into Sentence-like Units

Popis výsledku anglicky

This thesis deals with the problem of automatic segmentation of speech recognition output into sentence-like units. The work is focused on two languages - English and Czech. First, I describe creation of two Czech speech corpora with structural metadataannotation in two different domains: broadcast news and broadcast conversations. The main goal of this work is to develop automatic systems for dialog act segmentation of English multiparty meetings and sentence unit segmentation of the two new Czech corpora. I use and compare three modeling approaches - hidden Markov models, maximum entropy, and a boosting-based algorithm called BoosTexter. All of these approaches rely on two information sources - recognized words and prosody. In addition, I explore speaker adaptation for this task. The results indicate that superior performance is achieved when the three statistical models are combined via posterior probability interpolation.

Druh

O - Ostatní výsledky

CEP obor

JD - Využití počítačů, robotika a její aplikace

OECD FORD obor

—

Projekt

Výsledek vznikl pri realizaci vícero projektů. Více informací v záložce Projekty.

Návaznosti

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

Rok uplatnění

2008

Kód důvěrnosti údajů

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