Performance of Czech Speech Recognition with Language Models Created from Public Resources

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F11%3A00185981" target="_blank" >RIV/68407700:21230/11:00185981 - isvavai.cz</a>

Nalezeny alternativní kódy

RIV/46747885:24220/11:#0001963

Výsledek na webu

<a href="http://www.radioeng.cz" target="_blank" >http://www.radioeng.cz</a>

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

Performance of Czech Speech Recognition with Language Models Created from Public Resources

Popis výsledku v původním jazyce

In this paper, we investigate the usability of publicly available n-gram corpora for the creation of language models (LM) applicable for Czech speech recognition systems. N-gram LMs with various parameters and settings were created from two publicly available sets, Czech Web 1T 5-gram corpus provided by Google and 5-gram corpus created from the Czech National Corpus. We tested also a LM made of a large private resource of newspaper and broadcast texts collected by a Czech media mining company. The LMs were analyzed and compared via their perplexity rates and when employed in large vocabulary continuous speech recognition systems. Our study show that the Web1T-based LMs, even after intensive cleaning and normalization procedures, cannot compete with those made of smaller but more consistent corpora. The experiments done on large test data also illustrate the impact of Czech as highly inflective language on the perplexity, OOV, and recognition accuracy rates.

Název v anglickém jazyce

Performance of Czech Speech Recognition with Language Models Created from Public Resources

Popis výsledku anglicky

In this paper, we investigate the usability of publicly available n-gram corpora for the creation of language models (LM) applicable for Czech speech recognition systems. N-gram LMs with various parameters and settings were created from two publicly available sets, Czech Web 1T 5-gram corpus provided by Google and 5-gram corpus created from the Czech National Corpus. We tested also a LM made of a large private resource of newspaper and broadcast texts collected by a Czech media mining company. The LMs were analyzed and compared via their perplexity rates and when employed in large vocabulary continuous speech recognition systems. Our study show that the Web1T-based LMs, even after intensive cleaning and normalization procedures, cannot compete with those made of smaller but more consistent corpora. The experiments done on large test data also illustrate the impact of Czech as highly inflective language on the perplexity, OOV, and recognition accuracy rates.

Druh

J_x - Nezařazeno - Článek v odborném periodiku (Jimp, Jsc a Jost)

CEP obor

JA - Elektronika a optoelektronika, elektrotechnika

OECD FORD obor

—

Projekt

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

Návaznosti

Z - Vyzkumny zamer (s odkazem do CEZ)

Rok uplatnění

2011

Kód důvěrnosti údajů

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

Název periodika

Radioengineering

ISSN

1210-2512

e-ISSN

—

Svazek periodika

40

Číslo periodika v rámci svazku

4

Stát vydavatele periodika

CZ - Česká republika

Počet stran výsledku

7

Strana od-do

1002-1008

Kód UT WoS článku

000298636800039

EID výsledku v databázi Scopus

—