Voice query interfaces

For many users, voice-based interaction is the most natural form of communication. For this reason, we are working on voice interfaces for data analysis. This scenario creates interesting challenges when it comes to voice input as well as voice output.

Voice input can be hard to understand, compared to typed input, in particular in settings hwere background noise is prevalent. We have designed approaches that deal with uncertain query input by displaying results for different interpretations, optimizing the usage of screen space to maximize the probability of displaying the correct result.

Voice output is inherently limited in terms of the amount of information it can transmit. Un like text or visual output, users cannot scroll or skim output to gather the most relevant pieces of information quickly. Therefore, we developed approaches for summarizing query results optimally, based on the results of user studies evaluating limits in perception when listening to voice output. Our approaches analyze trends in query results and provide the best approximate voice description within length and time limits. At the same time, we interleave generation of voice output and data processing, to avoid processing steps that do not influence high-level data descriptions.