Using self-organizing maps to recognize acoustic units associated with information content in animal vocalizations
Citation
Placer, John, Slobodchikoff, Constantine N, Burns, Jason, Placer, Jeffrey, and Middleton, Ryan, Using self-organizing maps to recognize acoustic units associated with information content in animal vocalizations: .
Summary
Kohonenself-organizing neural networks, also called self-organizing maps (SOMs), have beenused successfully to recognize human phonemes and in this wayto aid in human speech recognition. This paper describes howSOMS also can be used to associate specific information contentwith animal vocalizations. A SOM was used to identify acousticunits in Gunnison's prairie dog alarm calls that were vocalizedin the presence of three different predator species. Some ofthese acoustic units and their combinations were found exclusively inthe alarm calls associated with a particular predator species andwere used to associate predator species information with individual alarmcalls. This methodology allowed individual alarm calls to be classifiedby predator [...]
Summary
Kohonenself-organizing neural networks, also called self-organizing maps (SOMs), have beenused successfully to recognize human phonemes and in this wayto aid in human speech recognition. This paper describes howSOMS also can be used to associate specific information contentwith animal vocalizations. A SOM was used to identify acousticunits in Gunnison's prairie dog alarm calls that were vocalizedin the presence of three different predator species. Some ofthese acoustic units and their combinations were found exclusively inthe alarm calls associated with a particular predator species andwere used to associate predator species information with individual alarmcalls. This methodology allowed individual alarm calls to be classifiedby predator species with an average of 91% accuracy. Furthermore,the topological structure of the SOM used in these experimentsprovided additional insights about the acoustic units and their combinationsthat were used to classify the target alarm calls. Animportant benefit of the methodology developed in this paper isthat it could be used to search for groups ofsounds associated with information content for any animal whose vocalizationsare composed of multiple simultaneous frequency components. �2006 Acoustical Society of America
Published in Journal of the Acoustical Society of America, volume 119, issue 5, on pages 3140 - 3146, in 2006.