Analysis of GlobalPhone and Ethiopian Languages Speech Corpora for Multilingual ASR

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4152–4156 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4152 Analysis of GlobalPhone and Ethiopian Languages Speech Corpora for Multilingual ASR Martha Yifiru Tachbelie1,2, Solomon Teferra Abate1,2, Tanja Schultz1 1Cognitive Systems Lab (CSL), University of Bremen, Germany 2School of Information Science, Addis Ababa University, Ethiopia {marthayifiru, abate, tanja.schultz}@uni-bremen.de Abstract In this paper, we present the analysis of GlobalPhone (GP) and speech corpora of Ethiopian languages (Amharic, Tigrigna, Oromo and Wolaytta). The aim of the analysis is to select speech data from GP for the development of multilingual Automatic Speech Recognition (ASR) system for the Ethiopian languages. To this end, phonetic overlaps among GP and Ethiopian languages have been analyzed. The result of our analysis shows that there is much phonetic overlap among Ethiopian languages although they are from three different language families. From GP, Turkish, Uyghur and Croatian are found to have much overlap with the Ethiopian languages. On the other hand, Korean has less phonetic overlap with the rest of the languages. Moreover, morphological complexity of the GP and Ethiopian languages, reflected by type to token ration (TTR) and out of vocabulary (OOV) rate, has been analyzed. Both metrics indicated the morphological complexity of the languages. Korean and Amharic have been identified as extremely morphologically complex compared to the other languages. Tigrigna, Russian, Turkish, Polish, etc. are also among the morphologically complex languages. Keywords: Language relatedness, Multilingual ASR, GlobalPhone, Ethiopian Languages 1. Introduction With more than 7000 languages in the world (Ethnologue, 2019) and the need to support multiple input and output languages, it is one of the most pressing challenge for the speech and language community to develop and deploy speech processing systems in yet unsupported languages rapidly and at reasonable costs (Schultz, 2004; Schultz and Kirchhoff, 2006). Major bottlenecks are the sparseness of speech and text data with corresponding pronunciation dic- tionaries, the lack of language conventions, and the gap between technology and language expertise. Data sparse- ness is a critical issue due to the fact that speech technolo- gies heavily rely on statistical modeling schemes, such as Hidden Markov Models, Deep Neural Networks (DNN) for acoustic modeling and n-gram and DNN for language mod- eling. Although statistical modeling algorithms are mostly language independent and proved to work well for a variety of languages, reliable parameter estimation requires vast amounts of training data. On the other hand, large-scale data resources for research are available for less than 100 languages and the costs for these collections are prohibitive to all but the most widely spoken and economically viable languages. This calls for the development of cross-lingual and/or multilingual speech processing/recognition systems. In cross-lingual ASR, resources of a language (source/donor language) are used to develop ASR system for another (target language) with or without little adaptation data from the target language. Multilingual ASR system is described as a system that is able to recognize multiple languages which are presented during training …

Content truncated to 500 words