Development of a conversational telephone speech recognizer for Levantine Arabic

Abstract

Manylanguages,includingArabic, are characterizedbyawidevariety of different dialectsthat often differ strongly from eachother. When developing speech technology for dialect-rich languages, the portability andreusability of data,algorithms, andsystemcomponents becomes extremely important. In this paper, we describe the development of a large-vocabulary speech recognition system for Levantine Arabic, which was a new dialectal recognition task for our existing system. We discuss the dialect-specific modeling choices (grapheme vs. phoneme based acoustic models, automatic vowelization techniques, and morphological languagemodels) and investigate to what extent techniquespreviously tested on other languagesare portable to the present task. We present stateof-the-art recognition results on the 2004 Levantine Arabic Rich Transcription evaluation.