Collecting Resources in Sub-Saharan African Languages for Automatic Speech Recognition: a Case Study of Wolof

Collecting Resources in Sub-Saharan African Languages for Automatic Speech Recognition: a Case Study of Wolof Elodie Gauthier1, Laurent Besacier1, Sylvie Voisin2, Michael Melese3, Uriel Pascal Elingui4 1Univ. Grenoble Alpes, LIG, F-38000 Grenoble, France 2Laboratoire Dynamique Du Langage (DDL), CNRS - Universit´e de Lyon, France 3Addis Ababa University, Ethiopia 4Voxygen SAS, Pleumeur-Bodou, France & Dakar, S´en´egal elodie.gauthier@imag.fr, laurent.besacier@imag.fr, sylvie.voisin@ish-lyon.cnrs.fr, michael.melese@gmail.com, elinguiuriel@voxygen.fr Abstract This article presents the data collected and ASR systems developped for 4 sub-saharan african languages (Swahili, Hausa, Amharic and Wolof). To illustrate our methodology, the focus is made on Wolof (a very under-resourced language) for which we designed the first ASR system ever built in this language. All data and scripts are available online on our github repository. Keywords: african languages, under-resourced languages, wolof, data collection, automatic speech recognition 1. Introduction Today is very favorable to the development of a market for speech in sub-saharan african languages. People’s access to information and communications technologies (ICT) is done mainly through mobile (and keyboard) and the need for voice services can be found in all sectors, from higher priority (health, food) to more fun (games, social media). For this, overcoming the language barrier is needed. This paper is done in the context of ALFFA project1 where two main aspects are involved: fundamentals of speech anal- ysis (language phonetic and linguistic description, dialec- tology) and speech technologies (automatic speech recog- nition (ASR) and text-to-speech (TTS)) for African lan- guages. In the project, developed ASR and TTS technolo- gies will be used to build micro speech services for mobile phones in Africa. For this, speech fundamental knowledge for targeted languages has to be upgraded while African language technologies are still at their very beginning. For these reasons, the ALFFA project is really interdisciplinary since it does not only gather technology experts but also in- cludes fieldwork linguists/phoneticians. Paper contribution. This article presents the data col- lected and ASR systems developped for 4 sub-saharan african languages (Swahili, Hausa, Amharic and Wolof). To illustrate our methodology, the focus is made on Wolof (a very under-resourced language) for which we designed the first ASR system ever built in this language. Paper outline. The outline of this paper is the follow- ing: first, section 2 will summarize ASR sytems already available on our github2 repository for Swahili, Hausa and Amharic (together with scripts and resources needed to re- produce ASR experiments). Then, sections 3, 4 and 5 will focus respectively on text data collection, speech data col- lection and ASR for Wolof language. Finally, section 6 will conclude this work and give some perspectives. 1see http://alffa.imag.fr 2see https://github.com/besacier/ALFFA_ PUBLIC/tree/master/ASR 2. ASR systems made available in the ALFFA project 2.1. Target languages of the project Language choice for the project is mainly governed by pop- ulation coverage and industrial perspectives. We focus on Hausa spoken by around 60 million people, as first or sec- ond language. Hausa is part of the family of Afroasiatic languages. Specifically, Hausa is the most spoken language among the Chadic languages (Vycichl, 1990). …

Content truncated to 500 words