We examine the role of transcription inconsistencies in the Faetar Automatic Speech Recognition benchmark, a challenging low-resource ASR benchmark. With the help of a small, hand-constructed lexicon, we conclude that find that, while inconsistencies do exist in the transcriptions, they are not the main challenge in the task. We also demonstrate that bigram word-based language modelling is of no added benefit, but that constraining decoding to a finite lexicon can be beneficial. The task remains extremely difficult.
@article{arxiv.2508.11771,
title = {Investigating Transcription Normalization in the Faetar ASR Benchmark},
author = {Leo Peckham and Michael Ong and Naomi Nagy and Ewan Dunbar},
journal= {arXiv preprint arXiv:2508.11771},
year = {2025}
}