| Popis: |
We introduce the discipline of Acoustic Geo-Sensing (AGS) that deals with the connection of acoustics and geoposition, i. e., `local audio' - focussing on spatial rather than on temporal aspects. We motivate this field of research, and give an example by automatic determination of a cyclist's route between determined start and endpoints, the direction she advances on this route, and the progress made from cell-phone audio. The Graz Cell-phone Cycle Corpus of 16 hours audio is introduced to this end. A standardised acoustic feature set ensures reproducibility throughout extensive experimentation aiming to reveal maximal spatiotemporal resolution. In the result, principle feasibility is shown by unsupervised clustering and all presented tasks can be solved at high accuracies and correlation within Random Forest classification and Additive Regression. |
