Experimentation on Spoken Format of Tables in Auditory User Interfaces

The acoustic representation of complex visual structures involves both synthesized speech and non-speech audio signals. Though progress in speech synthesis allows the consistent control of an abundance of parameters, like prosody through appropriate mark-up, there is not enough experimentally proven specification input data to drive a Voice Browser for such purposes. This paper reports on the results from a series of psychoacoustic experiments aiming to provide natural speech prosodic specification for the task of vocalizing tables. Blind and sighted listeners were asked to reconstruct simple and complex data tables from naturally spoken descriptions. From the listeners’ feedback it was deducted that consistent prosodic rendering can model the underlying semantic structure of tables.