ABSTRACT
Automated journalism is rapidly developing in the news industry. Among the most recent and promising technological potentials are neural voices, i.e., text-to-speech technology powered by neural networks. Based on a reception analysis with in-depth qualitative interviews (N = 12), this study explores how Danish radio listeners receive a full news broadcast read by a neural voice and perceive the credibility of the neural reader and the news content. Results show that the participants divide into two types: the perspicacious listeners who realize or suspect that the news reading is artificially synthesized and, to some degree, are annoyed by it, and the oblivious listeners who believe the news is read by a human and are predominantly positive towards it. Participants from both groups pay particular attention to voice emotionality when evaluating the appropriateness of the neural news reader. Also, they tend to attribute human characteristics to the neural news reader. The participants single out the news messages as well as the media organization behind the news broadcast, rather than the neural voice itself as critical components constituting credibility. Transparency is of great importance when applying a neural voice in a news broadcast, since it is a prerequisite for credibility.
Acknowledgements
We would like to thank Anders Kinch-Jensen, Filip Wallberg and Mads Vikkelsø, Center for Journalism, SDU for valuable contributions to this research project, and members of the Augmented Journalism network for feedback on an early version of this paper.
Disclosure Statement
No potential conflict of interest was reported by the author(s).
Notes
1 Meaning that the transcription, was cleaned up to remove filler words, stammers and anything that took away from the core message of what was being said.