By Boris Defréville, Pierre Roy, Christophe Rosin, and François Pachet.
120th Audio Engineering Society Convention, convention paper 6827, Paris, May 2006.
Read the paper (PDF)
AES convention record
About the paper
This paper describes an application of audio feature generation to urban sound monitoring. It studies how to identify common sound sources from recordings so that a monitoring system can account for what people hear in a given area.
Abstract summary
The authors use the Extractor Discovery System to identify task-specific acoustic features. They report preliminary results on six source categories and examine a two-stage classifier that first distinguishes vehicle sounds from other sources.
Related work
See Feature Generation for the method and The Bag-of-Frames Approach to Audio Pattern Recognition for related work comparing urban soundscapes and polyphonic music.