Samrómur: Crowd-sourcing Data Collection for Icelandic Speech Recognition

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3463–3467 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3463 Samrómur: Crowd-sourcing Data Collection for Icelandic Speech Recognition David Erik Mollberg, Ólafur Helgi Jónsson, Sunneva Þorsteinsdóttir, Steinþór Steingrímsson, Eydís Huld Magnúsdóttir, Jón Guðnason Reykjavik University Menntavegi 1, 102 Reykjavik {de14, olafurhj, sunnevath, steinthor18, eydishm, jg}@ru.is Abstract This contribution describes an ongoing project of speech data collection, using the web application Samrómur which is built upon Common Voice, Mozilla Foundation’s web platform for open-source voice collection. The goal of the project is to build a large-scale speech corpus for Automatic Speech Recognition (ASR) for Icelandic. Upon completion, Samrómur will be the largest open speech corpus for Icelandic collected from the public domain. We discuss the methods used for the crowd-sourcing effort and show the importance of marketing and good media coverage when launching a crowd-sourcing campaign. Preliminary results exceed our expectations, and in one month we collected data that we had estimated would take three months to obtain. Furthermore, our initial dataset of around 45 thousand utterances has good demographic coverage, is gender-balanced and with proper age distribution. We also report on the task of validating the recordings, which we have not promoted, but have had numerous hours invested by volunteers. Keywords: Speech corpora, Icelandic, Crowd-sourcing 1. Introduction We describe a work in progress aimed at collecting speech data employing crowd-sourcing. The work started in mid- 2019 and a crowd-sourcing platform was launched in fall the same year. It has been very well received and by the end of the first month we had surpassed the data collection goals set for the first three months of the project. Rapid development in recent years in voice-controlled con- sumer products has made the apparent stagnation of such solutions for Icelandic all the more noticeable. This project started out as an effort to translate the Com- mon Voice platform1 to Icelandic, in collaboration with De- loitte Iceland, a private company. The effort soon got in- corporated into a larger program, a new national program for language technology (LT), a five-year national LT pro- gram launched in October (Nikulásdóttir et al., 2020). It is funded by the government, managed by Almannarómur, an NGO, and executed by a consortium of universities, re- search institutes and private enterprise. From the start all facets of this crowd-sourcing effort to collect speech data have been collaborative, which has been a vital part of its success so far. A report (Nikulásdóttir et al., 2017), written at the prepa- ration stage for the national LT program, describes neces- sary actions and requirements for building a well working LT infrastructure in Iceland and sets out goals for various core projects of the program. One of these core projects concerns building better speech recognition systems. To this end, goals are set for collecting speech data for train- ing. The goals proposed in the report include collecting 200,000 utterances from adult speakers, 200,000 utterances from youth …

Content truncated to 500 words