Developing further speech recognition resources for Welsh

Abstract

This paper reports on ongoing research into developing large-vocabulary continuous speech recognition (LVCSR) for the Welsh language. We address data design issues and the method for data collection using a purposely designed application for mobile devices. We also discuss the application of the data including the design and collection of a small speech corpus to cover the commands used to control a robotic arm in Welsh on a Raspberry Pi computer the licensing of the project and our hopes for the application of the project resources to other languages.

Crowdsourcing the Paldaruo Speech Corpus of Welsh for Speech Technology

“…The results of this project will be available in early 2020. As part of the GALLU project, the speech data was successfully used to build acoustic models that recognized commands to move a robot arm connected to a Raspberry Pi using spoken commands in Welsh [42]. As part of subsequent projects, packages for speech recognition have been built in Docker-based environments for users to easily produce, test and apply acoustic models in Kaldi [40], HTK [39] and Julius [43] using the Paldaruo Corpus.…”