Selezione lingua
ItalianoInglese
Cerca nel sito
Area riservata
Tot. visite contenuti : 342624
Home Bollettino Tecnologie del linguaggio 2013-004T - Comparison of Grapheme-to-Phoneme Methods on Large Pronunciation Dictionaries and LVCSR TasksComparison of Grapheme-to-Phoneme Methods

PostHeaderIcon 2013-004T - Comparison of Grapheme-to-Phoneme Methods on Large Pronunciation Dictionaries and LVCSR TasksComparison of Grapheme-to-Phoneme Methods

Stefan Hahn, Paul Vozila, Maximilian Bisani, Comparison of Grapheme-to-Phoneme Methods on Large Pronunciation Dictionaries and LVCSR TasksComparison of Grapheme-to-Phoneme Methods

DOI: --

URL: http://www-i6.informatik.rwth-aachen.de/publications/download/811/Hahn--2012.pdf

Parole chiave: grapheme-to-phoneme conversion, G2P, ASR

Abstract: Grapheme-to-Phoneme conversion (G2P) is usually used within every state-of-the-art ASR system to generalize beyond a fixed set of words. Although the performance is typically already quite good (< 10% phoneme error rate) and pronunciations of important words are checked by a linguist, further improvements are still desirable, especially for end user customization. In this work, we present and compare five methods/tools to tackle the G2P task. Although most of the methods have already been published and/or are available as open source software, the reported experiments are done on large state-of-the-art tasks and the used software is from the actual publications. Besides an experimental comparison on text data for a range of languages (i.e. measuring the G2P accuracy only), our focus in this paper is measuring the effect of improved G2P modeling on LVCSR performance for a challenging ASR task. Additionally, the effect of using n-Best pronunciation variants instead of single best is investigated briefly.

Commento: Up to date comparison of state-of-the-art Grapheme-to-Phoneme methods with a focus on LVCSR.

Ultimo aggiornamento ( Mercoledì 24 Aprile 2013 11:45 )