Forced Alignment for Understudied Language Varieties: Testing Prosodylab-Aligner with Tongan Data

Abstract

Automated alignment of transcriptions to audio files expedites the process of preparing data for acoustic analysis. Unfortunately, the benefits of auto-alignment have generally been available only to researchers studying majority languages, for which large corpora exist and for which acoustic models have been created by large-scale research projects. Prosodylab-Aligner (PL-A), from McGill University, facilitates automated alignment and segmentation for understudied languages. It allows researchers to train acoustic models using the same audio files for which alignments will be created. Those models can then be used to create time-aligned Praat TextGrids with word and phone boundaries marked.

For the benefit of others who wish to use PL-A for research projects, this paper reports on our use of PL-A on Tongan field recordings, reviewing the software, outlining required steps, and providing tips. Since field recordings often contain more background noise than the laboratory recordings for which PL-A was designed, the paper also discusses the relative benefits of removing background noise for both training and alignment purposes. Finally, it compares acoustic measures based on various alignments and compares boundary placements with those of human aligners, demonstrating that automated alignment is both feasible and less time-consuming than manual alignment.