OOV detection in LVCSR using neural networks

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26230%2F08%3APU82717" target="_blank" >RIV/00216305:26230/08:PU82717 - isvavai.cz</a>

Result on the web

—

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

OOV detection in LVCSR using neural networks

Original language description

Confidence measures and classifying techniques are widely used for the recognition error detection task in LVCSR (Large Vocabulary Continuous Speech Recognition). But in many recognition scenarios the amount of words not included in the dictionary (e.g.real names, neologisms) lead to so-called OOV (Out Of Vocabulary) errors which increase the WER (Word Error Rate) even more. The hereby described work acknowledges and investigates further improvements of an OOV detection task performed by combining strong and weak phone posterior features using neural networks based on [ICASSP08] and the use of phone context.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

JC - Computer hardware and software

OECD FORD branch

—

Project

—

Continuities

Z - Vyzkumny zamer (s odkazem do CEZ)

Publication year

2008

Confidentiality

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

Article name in the collection

Proc. STUDENT EEICT 2008

ISBN

978-80-214-3617-6

ISSN

—

e-ISSN

—

Number of pages

3

Pages from-to

—

Publisher name

Faculty of Electrical Engineering and Communication BUT

Place of publication

Brno

Event location

FEKT VUT v Brně

Event date

Apr 24, 2008

Type of event by nationality

CST - Celostátní akce

UT code for WoS article

—