Speech Corpus of Ainu Folklore and End-to-end Speech Recognition for Ainu Language

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2622–2628 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2622 Speech Corpus of Ainu Folklore and End-to-end Speech Recognition for Ainu Language Kohei Matsuura, Sei Ueno, Masato Mimura, Shinsuke Sakai, Tatsuya Kawahara Graduate School of Informatics, Kyoto University Sakyo-ku, Kyoto 606-8501, Japan {matsuura, ueno, mimura, sakai, kawahara}@sap.ist.i.kyoto-u.ac.jp Abstract Ainu is an unwritten language that has been spoken by Ainu people who are one of the ethnic groups in Japan. It is recognized as critically endangered by UNESCO and archiving and documentation of its language heritage is of paramount importance. Although a considerable amount of voice recordings of Ainu folklore has been produced and accumulated to save their culture, only a quite limited parts of them are transcribed so far. Thus, we started a project of automatic speech recognition (ASR) for the Ainu language in order to contribute to the development of annotated language archives. In this paper, we report speech corpus development and the structure and performance of end-to-end ASR for Ainu. We investigated four modeling units (phone, syllable, word piece, and word) and found that the syllable-based model performed best in terms of both word and phone recognition accuracy, which were about 60% and over 85% respectively in speaker-open condition. Furthermore, word and phone accuracy of 80% and 90% has been achieved in a speaker-closed setting. We also found out that a multilingual ASR training with additional speech corpora of English and Japanese further improves the speaker-open test accuracy. Keywords: Ainu speech corpus, low-resource language, end-to-end speech recognition 1. Introduction Automatic speech recognition (ASR) technology has been made a dramatic progress and is currently brought to a prat- ical levels of performance assisted by large speech corpora and the introduction of deep learning techniques. However, this is not the case for low-resource languages which do not have large corpora like English and Japanese have. There are about 5,000 languages in the world over half of which are faced with the danger of extinction. Therefore, con- structing ASR systems for these endangered languages is an important issue. The Ainu are an indigenous people of northern Japan and Sakhakin in Russia, but their language has been fading away ever since the Meiji Restoration and Modernization. On the other hand, active efforts to preserve their culture have been initiated by the Government of Japan, and excep- tionally large oral recordings have been made. Neverthe- less, a majority of the recordings have not been transcribed and utilized effectively. Since transcribing them requires expertise in the Ainu language, not so many people are able to work on this task. Hence, there is a strong demand for an ASR system for the Ainu language. We started a project of Ainu ASR and this article is the first report of this project. We have built an Ainu speech corpus based on data pro- vided by the Ainu Museum1 and the Nibutani Ainu Cul- ture Museum2. The oral recordings …

Content truncated to 500 words