Introducing Bed Word: a new automated speech recognition tool for sociolinguistic interview transcription

Abstract

Abstract We present Bed Word, a tool leveraging industrial automatic speech recognition (ASR) to transcribe sociophonetic data. While we find lower accuracy for minoritized English varieties, the resulting vowel measurements are overall very close to those derived from human-corrected gold data, so fully automated transcription may be suitable for some research purposes. For purposes requiring greater accuracy, we present a pipeline for human post-editing of automatically generated drafts, which we show is far faster than transcribing from scratch. Thus, we offer two ways to leverage ASR in sociolinguistic research: full automation and human post-editing. Augmenting the DARLA tool developed by Reddy and Stanford (2015b. Toward completely automated vowel extraction: Introducing DARLA. Linguistics Vanguard 1(1). 15–28), we hope that this resource can help speed up transcription for sociophonetic research.