Unified Approach to Development of ASR Systems for East Slavic Languages

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F46747885%3A24220%2F17%3A00004820" target="_blank" >RIV/46747885:24220/17:00004820 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1007/978-3-319-68456-7_16" target="_blank" >http://dx.doi.org/10.1007/978-3-319-68456-7_16</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-319-68456-7_16" target="_blank" >10.1007/978-3-319-68456-7_16</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Unified Approach to Development of ASR Systems for East Slavic Languages

Popis výsledku v původním jazyce

This paper deals with the development of language specific modules (lexicons, phonetic inventories, LMs and AMs) for Russian, Ukrainian and Belarusian (used by 260M, 45M and 3M native speakers, respectively). Instead of working on each language separately, we adopt a common approach that allows us to share data and tools, yet taking into account language unique features. We utilize only freely available text and audio data that can be found on web pages of major newspaper and broadcast publishers. This must be done with large care, as the 3 languages are often mixed in spoken and written media. So, one component of the automated training process is a language identification module. At the output of the complete process there are 3 pronunciation lexicons (each about 300K words), 3 partly shared phoneme sets, and corresponding acoustic (DNN) and language (N-gram) models. We employ them in our media monitoring system and provide results achieved on a test set made of several complete TV news in all the 3 languages. The WER values vary in range from 24 to 36%.

Název v anglickém jazyce

Unified Approach to Development of ASR Systems for East Slavic Languages

Popis výsledku anglicky

This paper deals with the development of language specific modules (lexicons, phonetic inventories, LMs and AMs) for Russian, Ukrainian and Belarusian (used by 260M, 45M and 3M native speakers, respectively). Instead of working on each language separately, we adopt a common approach that allows us to share data and tools, yet taking into account language unique features. We utilize only freely available text and audio data that can be found on web pages of major newspaper and broadcast publishers. This must be done with large care, as the 3 languages are often mixed in spoken and written media. So, one component of the automated training process is a language identification module. At the output of the complete process there are 3 pronunciation lexicons (each about 300K words), 3 partly shared phoneme sets, and corresponding acoustic (DNN) and language (N-gram) models. We employ them in our media monitoring system and provide results achieved on a test set made of several complete TV news in all the 3 languages. The WER values vary in range from 24 to 36%.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

20204 - Robotics and automatic control

Projekt

<a href="/cs/project/TA04010199" target="_blank" >TA04010199: MULTILINMEDIA - Multilinguální platforma pro monitoring a analýzu multimédií</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)<br>I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2017

Kód důvěrnosti údajů

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

Název statě ve sborníku

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ISBN

9783319684550

ISSN

0302-9743

e-ISSN

—

Počet stran výsledku

11

Strana od-do

193-203

Název nakladatele

Springer Verlag

Místo vydání

Německo

Místo konání akce

Le Mans, Francie

Datum konání akce

1. 1. 2017

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—