VoicePredict – Understanding movement through location and speech information
A GPS point describes a location, but not an intention. VoicePredict investigated what additional information audio and context can contribute to predicting movement. Its research question concerned where a person or vehicle was likely to be after a given time interval.
Location histories and audio were treated as distinct sources. They have to align in time, provide useful features and be processed in a joint prediction model. The work investigated machine-learning methods and statistical models for connecting that information.
© FILA
The project worked with a regional industrial partner and transferred the developed software to that partner. Evaluation depends on the dataset, prediction horizon and evaluation method: an additional information source is useful when its contribution can be tested under defined conditions.
VoicePredict is a completed project in the background of current FILA multimodal research. AURORA continues investigating GPS histories and speech context; AI and multimodal analysis connects the question with current datasets and publications.
Project and funding
Status: completed; funding phase 2020–2022. Funding: ERDF / the state of Saxony-Anhalt.
