Samrómur Children: An Icelandic Speech Corpus

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 995–1002 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 995 Samr´omur Children: An Icelandic Speech Corpus Carlos Mena, David Erik Mollberg, Michal Borsk´y, J´on Guðnason Language and Voice Lab, Reykjav´ık University Menntavegur 1, 102 Reykjav´ık, Iceland {carlosm, de14 ,michalb, jg}@ru.is Abstract Samr´omur Children is an Icelandic speech corpus intended for the field of automatic speech recognition. It contains 131 hours of read speech from Icelandic children aged between 4 to 17 years. The test portion was meticulously selected to cover a wide range of ages as possible as we aimed to have exactly the same amount of data per age range. The speech was collected with the crowd-sourcing platform samromur.is, which is inspired on the “Mozilla’s Common Voice Project”. The corpus was developed within the framework of the “Language Technology Programme for Icelandic 2019 − 2023”; the goal of the project is to make Icelandic available in language-technology applications. Samr´omur Children is the first corpus in Icelandic with children’s voices for public use under a Creative Commons license. Additionally, we present baseline experiments and results using Kaldi. Keywords: children’s speech corpus, children’s speech recognition, icelandic children’s speech, icelandic corpus. 1. Introduction 1.1. Project Background The creation of “Samr´omur Children” is part of the “Language Technology Programme for Icelandic 2019 − 2023” (Nikul´asd´ottir et al., 2020) which aims to make Ice- landic available in language-technology applications, as well as in all areas of communication, with all the deliv- erables of the programme published under open licenses to encourage the use of them in commercial products such as pronunciation apps to help children to improve their read- out-loud skills or to learn Icelandic. Samr´omur Children is a subset of the data gathered by crowd-sourcing using the web application “Samr´omur” (Mollberg et al., 2020), an ongoing project of speech data collection, based on the “Mozilla’s Common Voice Project” for open-source voice collection (Ardila et al., 2019). The goal of Samr´omur is to build a large-scale speech corpus for Automatic Speech Recognition (ASR) for Icelandic as it is contemplated in the programme. It was therefore necessary to set up a sepa- rate platform from the Mozilla’s one, in order to have more control over the data gathering and the distribution of the datasets when it comes to consent forms and meta-data col- lected. Samr´omur is the largest prompted speech collection ef- fort for Icelandic so far. However, previous efforts in this respect have been deployed before. One exam- ple of this, is the “Icelandic Speech Recognition Project Hjal” (R¨ognvaldsson, 2003) with the main goal of collect- ing sufficient material to train a speaker-independent iso- lated word recognition system. Another example is the “Almannar´omur Project” (Guðnason et al., 2012), started as a collaboration between the Reykjavik University and the Icelandic Centre for Language Technology in cooper- ation with Google, that provided technical support. One of the main products delivered by the Almannar´omur Project is the M´alr´omur corpus …

Content truncated to 500 words