Investigating Transcription Normalization in the Faetar ASR Benchmark

TL;DR

This paper investigates transcription inconsistencies in the Faetar ASR benchmark, finding they are not the main challenge, and explores language modeling and decoding constraints to improve low-resource speech recognition.

Contribution

It provides an analysis of transcription issues in Faetar ASR and evaluates the impact of language modeling and lexicon constraints on performance.

Findings

Transcription inconsistencies are not the primary challenge.

Bigram language models do not improve performance.

Lexicon-constrained decoding can be beneficial.

Abstract

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.