Huqariq: A Multilingual Speech Corpus of Native Languages of Peru forSpeech Recognition

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5029–5034 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5029 Huqariq: A Multilingual Speech Corpus of Native Languages of Peru for Speech Recognition Rodolfo Zevallos, Luis Camacho, Nelsi Melgarejo Pompeu Fabra University, Pontifical Catholic University of Peru Barcelona Spain, Lima Peru rodolfojoel.zevallos@upf.edu, {luis.camacho, nelsi.melgarejo}@pucp.pe Abstract The Huqariq corpus is a multilingual collection of speech from native Peruvian languages. The transcribed corpus is intended for the research and development of speech technologies to preserve endangered languages in Peru. Huqariq is primarily designed for the development of automatic speech recognition, language identification and text-to-speech tools. In order to achieve corpus collection sustainably, we employ the crowdsourcing methodology. Huqariq includes four native languages of Peru, and it is expected that by the end of the year 2022, it can reach up to 20 native languages out of the 48 native languages in Peru. The corpus has 220 hours of transcribed audio recorded by more than 500 volunteers, making it the largest speech corpus for native languages in Peru. In order to verify the quality of the corpus, we present speech recognition experiments using 220 hours of fully transcribed audio. Keywords: Speech Corpus, Speech Recognition, Low-resource Languages 1. Introduction The Huqariq project responds to the endangerment currently faced by native languages in Latin Amer- ica and the lack of language technologies faced by low-resource languages in Peru (Rogers and Camp- bell, 2015). This situation is mainly due to the lack of speech corpora, which are the raw material for the creation of language tools, are scarce and the few that exist are privately licensed; for this reason, they should be in the public domain to contribute to the develop- ment and revitalization of languages. Around the world, there are some initiatives for the col- lection of corpora for low-resources languages and that are in the public domain that employ different method- ologies and ways of collection. One of the most suc- cessful methodologies is crowdsourcing, i.e. native speakers volunteering to help in the construction of the resources. This methodology is supported by web tools or mobile applications, which can be massively used. Our corpus collection tool is designed to expand or- ganically to new native languages as community mem- bers record domain-specific base audios as prompts in the corpus collection. Unlike others (Common Voice (Ardila et al., 2019)), this tool does not use text to be read, as many native speakers of indigenous lan- guages are illiterate in their native language. Therefore, our tool replaces reading texts with listening to audios. This subtle but important change facilitates corpus col- lection. 2. Prior work Although the majority of the speech corpora employed in the most widely used tools are private, there are some worldwide initiatives of speech corpora with open li- censes and low-resource languages. In 2019 the Vox- Forge project (VoxForge, 2019) collected a speech cor- pus for 17 languages; this project is community-driven in the …

Content truncated to 500 words