Combining Text-to-Speech Services with Conventional Voiceover for News Oralization

Abstract

The surge in digital content consumption has, in many cases, posed challenges for media companies, resulting in reduced revenue and the need to reinvent business models. The digitalization of content has introduced new consumption formats, and news podcasts have already become a reality in this landscape. While their existence is relatively recent in journalism, the increasing popularity of this format makes it an appealing addition to the field. But the production of podcasts may be demanding in what relates to the time, resources and even the technical expertise needed. In this scope, this paper primarily focuses on the premise of facilitating the creation of news podcasts. To achieve this, we propose employing Text-to-Speech technologies (TTS) for the oralization of journalistic texts in European Portuguese. We conducted tests using TTS services from Amazon Polly and Google Speech Cloud, with Google Speech Cloud Wavenet services yielding superior results among potential users. Additionally, we developed three podcast models incorporating human voiceover and/or TTS to get the users acceptance of those models. One model used only human voices, another only voice created with TTS and a hybrid podcast integrating both types of voices. The presence of human voice positively influenced the results, with the human voice model and hybrid voice outperforming the exclusive TTS voice model. However, the differences between the models were not significantly pronounced, and the results demonstrated an acceptance of Text-to-Speech technology in the context of news podcasts. Nonetheless, there remains a need for continuous technological advancement to converge with human discourse.