Acoustic environments, such as street, office and supermarket, provide a rich source of information on the types of activity, communication modes and other actors involved. This project explored acoustic environments as a source of context information, its capture using consumer recording devices and its classification in context-aware information systems.
Our work showed that acoustic environments can be accurately classified using sound samples acquired from widely available consumer devices. Classification using HMMs and MFCCs are adequate for building an acoustic environment classifier and yield good results. The overall accuracy in high bandwidth classification is 92.27% and achieved 95.83% in the low bandwidth classification. The classifier also performed well in recognising single source sound events. A context tracker has been developed to show the feasibility of capturing and recognising acoustic environment using mobile devices and in low-bandwidth communication environments.We believe that an adaptive learning strategy based on confidence measurements and a hierarchical model of acoustic environments can further improve the recognition accuracy. However, our work with the RWCP and other data suggests that constructing a useful and robust hierarchical model of sounds will not be easy.
This is the data used in:
A zip archive of spectrograms for the low bitrate data are available in .eps and Matlab .fig forms
The data we collected is available as a set of zip files, varying in size between about 2MB and 10MB each.
| High bitrate data (.wav) |
Low bitrate data (.wav) |
Low bitrate data (MFCC) |
|---|---|---|
| beach | 10secsamples | 10secsamples |
| bus | buildingsite | buildingsite |
| car | bus | bus |
| city | car | car |
| football | car_high | car_high |
| laundrette | laundrette | laundrette |
| office | office | office |
| lecture | presentation | presentation |
| pub | shoppingcentre | shoppingcentre |
| railstation | street_people | street_people |
| silence | street_traffic | street_traffic |
| street | train | train |