Augmented Journalism webinar: Presentation on The Robot as Newsreader: Exploring the attitudes of listeners to automated text reading in radio news broadcast

Aktivitet: Foredrag og mundtlige bidragKonferenceoplæg


Automated text reading of news is currently applied on many webpages as a tool for users who need – or prefer – to listen to the news, rather than reading it. Now, news organizations have begun to explore further possibilities for applying text-to-speech (TTS) technology in journalism. With the latest implementation of neural networks, TTS technology is making such rapid progress in terms of natural pronunciation and intonation that the technology has the potential to supplement regular human news reading on radio, in podcasts as well as on TV as voice-overs. The purpose of this study is to examine how radio listeners experience news reading by speech synthesizers, i.e. TTS, powered by deep neural networks. Recent research in other fields has shown that neural TTS voices are rated as more human-like, natural, familiar and likeable than traditional TTS voices (Cohn & Zellou 2020). Still, journalism research has paid limited attention to the potentials and pitfalls of automatic text-reading (Scott, Ashby & Cibin, 2020). Following other journalistic research into news automation (e.g. Clerwall, 2014; Thurman et al. 2018), we argue that the audience reaction to this innovation will determine its implementation in the newsroom. Research question While the scenario of “robots” reading the news may be deemed dystopian science fiction by some, the progressive use of algorithms in written news journalism has already proven that automated journalism is in demand in the media industry (Graefe 2016), for example in news about sports results and business. In addition, research shows that computer-generated news is perceived by the readers as credible and in many regards similar to regular journalism (Graefe et al. 2016, Wölker & Powell 2018). As in the case of machine-written news, the automation of news reading might be considered a low-cost opportunity for journalists to allocate time from routine tasks to in-depth reporting instead (cf. van Dalen 2012). However, from the perspective of the users, there might be an essential difference between robots writing the news and robots reading news. People use different criteria to evaluate the trustworthiness of written and broadcast news. Newhagen and Nass (1989) showed that the personality of journalists has a much stronger effect on the credibility of television news than for confidence in newspapers. This aligns with the observation by Aristotle that ethos is an essential source of a communicator’s credibility. Since “voice is such a personal thing” (Skyum-Nielsen 2008: 409), it may, therefore, not necessarily be suited for mechanical renditions in broadcast journalism. Automated news-reading might undermine the importance of the journalist’s voice as an essential source of credibility. This beckons our exploratory research question: RQ: How do listeners experience news reading in the radio read by text-to-speech technology, and which attitudes do they assume towards the naturalness and trustworthiness of the voice and the – potentially – perceived persona behind it? Specifically, we investigate the neural voice “Christel”, who speaks Danish and has been recently developed by Microsoft as a part of their cognitive speech services and Azure package, which is considered cutting-edge technology within the field.
Periode28. jan. 2021
Sted for afholdelseAugmented Journalism webinar
Grad af anerkendelseInternational