Less-resourced languages face additional challenges in the creation of tools and resources for speech recognition applications. These include lack of funding, sparsity of data and shortage of experts with relevant skills. On the other hand there are also opportunities to be had from tapping into committed communities of language activists, and potentially developing innovative solutions to common problems that may be applied elsewhere. This paper describes a recent series of short-term projects for the Welsh language that have used crowdsourcing methodologies, together with data from Wicipedia (the Welsh Wikipedia) and existing Welsh corpora, to further advance the field. They have also borrowed and adapted open source tools, such as MaryTTS and Mozilla CommonVoice that were already freely available. In addition this paper provides some pointers towards further needs and solutions for speech technology in less-resourced languages, aiming at a coherent, long-term approach that may be applicable in many environments.
@InProceedings{PRYS18.6, author = {Delyth Prys and Dewi Jones}, title = {Gathering Data for Speech Technology in Welsh: A Case Study}, booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)}, year = {2018}, month = {may}, date = {7-12}, location = {Miyazaki, Japan}, editor = {Claudia Soria and Laurent Besacier and Laurette Pretorius}, publisher = {European Language Resources Association (ELRA)}, address = {Paris, France}, isbn = {979-10-95546-22-1}, language = {english} }
Summary of the paper