Evaluating speech synthesis in a mobile context: Audio presentation of Facebook, Twitter and RSS

摘要

This paper presents an evaluation of a podcast service that aggregates data from Facebook, Twitter and RSS feeds, using speech synthesis. The service uses a novel approach to speech synthesis generation, where XML markup is used to control both the speech synthesis and the sound design of a resulting podcast. A two-phase evaluation was carried out: 1) participants listening to the podcasts on desktop computers, 2) participants listening to the podcasts while walking. Our findings show that participants preferred shorter podcasts with sound effects and background music, and were affected by the surrounding environmental noise. However, audio advertising which is part of the service did not have a significant negative effect. Another finding was that the advantage of using multiple voices for content segmentation may have been undermined by difficulties in listener adaptation. The work is part of a new approach to speech synthesis provision, where its style of rendition forms a part of the application design and it is evaluated within an application context.