A Romanization System and WebMAUS Aligner for Arabic Varieties

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 7269–7276 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 7269 A Romanization System and WebMAUS Aligner for Arabic Varieties Jalal Al-Tamimi∗, Florian Schiel∗∗, Ghada Khattab†, Navdeep Sokhey‡‡, Djegdjiga Amazouz‡, Abdulrahman Dallak††, Hajar Moussa⋆ ∗Université Paris Cité, CNRS, Laboratoire de linguistique formelle (LLF), F-75013 Paris France ∗∗Bavarian Archive for Speech Signals, Schellingstr. 3 80799 München Germany †Newcastle University, Newcastle upon Tyne, NE1 7RU UK ‡‡Virginia Polytechnic Institute and State University, 220 Stanger Street, 331 Major Williams Hall (0225), ‡‡Laboratoire de Phonétique et Phonologie (LPP)-CNRS, Université Sorbonne Nouvelle, 75005, Paris France ††Newcastle University, Newcastle upon Tyne, NE1 7RU UK and Jazan University, Jazan, 45142, Saudi Arabia, ⋆King Abdul-Aziz University, Jeddah, 21589, Saudi Arabia. ∗jalal.al-tamimi@u-paris.fr, ∗∗schiel@bas.uni-muenchen.de, †ghada.khattab@newcastle.ac.uk, ‡‡navsokhey@vt.edu, ‡djegdjiga.amazouz@sorbonne-nouvelle.fr, ††A.Dallak2@newcastle.ac.uk/††adallak@jazanu.edu.sa, ⋆hmoussa@kau.edu.sa Abstract This paper presents the results of an ongoing collaboration to develop an Arabic variety-independent romanization system that aims to homogenize and simplify the romanization of the Arabic script, and introduces an Arabic variety-independent WebMAUS service offering a free to use forced-alignment service fully integrated within the WebMAUS services. We present the rationale for developing such a system, highlighting the need for a detailed romanization system with graphemes corresponding to the phonemic short and long vowels/consonants in Arabic varieties. We describe how the acoustic model was created, followed by several hands-on recipes for applying the forced alignment webservice either online or programatically. Finally, we discuss some of the issues we faced during the development of the system. Keywords: Arabic, romanization, dialect-independent, WebMAUS, aligner 1. Introduction and Motivation The WebMAUS1 is a suite of webservices that is free for academic users that – at the time of writ- ing – comprises 22 speech and language processing tools (Kisler et al., 2017). Since its introduction in 2013, roughly 17 million media files from users from all over the world have been processed by this service suite. The core component around which most of the services are grouped is the well known WebMAUS speech segmentation and labelling en- gine first introduced in 1999 (Schiel, 1999). A pow- erful pipeline framework allows the concatenation of several individual services; for instance, auto- matic phonetic and syllabic segmentation and la- belling of a speech recording is first performed us- ing Automatic Speech Recognition (ASR) followed by a Text-to-phoneme translation, the WebMAUS engine and a Syllabification service in one process- ing call. Another point that distinguishes the Web- MAUS services is the fact that users can choose be- tween a sophisticated web interface or a program- matic access to each service (and the pipeline) us- ing REST based synchronous service calls2. 1https://hdl.handle.net/11858/00-1779-000 0-0028-421B-4, also known as “BAS WebServices” 2See https://clarin.phonetik.uni-muenchen.de /BASWebServices/services/ help for details Currently, the WebMAUS services process 42 lan- guages and language varieties, and utilize 1243 trained acoustic phone models based on HTK (Young et al., 2015). Up to now, Arabic varieties have not been part of the supported languages. The reasons are two-fold: firstly, there is a gen- …

Content truncated to 500 words