Speech-To-Local Data: Exploring ASR Files of Archived Television News (2004-2018) on the 1986 Chernobyl Nuclear Disaster

Abstract

Broadcast archives have started to use Automatic Speech Recognition (ASR) to generate transcripts of their audio-visual collections to open up the latter to media professionals and media researchers. In this methodological article, we study the ASR files of Dutch archived television news on the 1986 Chernobyl nuclear disaster. By unpacking ASR's specificities and exploring methodologies, we examine the repackaging of the Chernobyl nuclear disaster on Dutch television. Building upon Loukissas' concept of 'local data', combined with insights from archival studies, digital history, and television studies, we argue that ASR ought to be considered, not just as speech-to-text, but also as speech-to-local data in which the production contexts of broadcast archives and of television newsrooms need to be taken into account. ASR appears to be a specific kind of repackaging in which the broadcasts are segmented, programmed, and archived, and subsequently used to generate a time code and speech transcript, including errors, gaps, schemata, and traces of rituals that - together - are shaping and reshaping the historical record, in this case of the 1986 Chernobyl nuclear disaster.