Resources and tools for pre-processing speech data in a lesser-known variety of English

Abstract

Research on lesser-known language varieties can be hindered from the outset by the need for both data and tools for automating the required pre-processing work. For speech, whilst more ecologically valid data in the form of video and audio are sometimes available, these need to be accompanied by a machine-readable text, ideally segmented and labelled, both allowing for searchability. A significant initial commitment is needed even before the relevant phonetic and phonological research can begin. This paper demonstrates and evaluates the efficacy of already existing tools (YouTube captioning and WebMAUS forced alignment) in automating the pre-processing work required using Maltese English (MaltE), whilst also showcasing a sample analysis of the pronunciation of post-vocalic ‘r’ in the variety. As a low resource variety of English, MaltE presents a test case for showing how existing resources and tools can be utilised to work with language varieties which are digitally less well-supported.