we made binaural recordings of 25 environmental sounds, which were then
played back to a jury via headphones. The jury had the task to judge upon
the sounds by means of several semantic differentials. By factor analysis of
the subject's answers 6 invariant principal dimensions were extracted, which
can be used to differenciate between the sounds.
Whereas the above analysis accounts for the description of the judgements,
however, I trained neural networks to do the same task, namely to map the
sound recordings onto the dimensions extracted from the test. The temporal
extension to the self-organized-feature-map produced the best results: the
sounds were clustered according to their spectral and short-term temporal
features. This way, sounds with tonal, noisy, stationary or fluctuating
features were arranged on a two dimensional map.
The work is described in my Ph.D.-thesis, however you can find the feature
map results also in
Prante, Holger U. & Koop, Lars; Classification of Sounds with Temporal
Feature Maps. In: Schick A., Meis, M., Reckhardt, C. (eds.) 2000,
Contributions to Psychological Acoustics. Results of the 8th Oldenburg
Symposium on Psychological Acoustics. Oldenburg University Press.
Holger U. Prante
Institute of Technical Acoutics
Technical University of Berlin
Von: Pawel Kusmierek <pq@NENCKI.GOV.PL>
An: AUDITORY@LISTS.MCGILL.CA <AUDITORY@LISTS.MCGILL.CA>
Datum: 20 March, 2001 2:34 PM
Betreff: How to classify/categorize sounds
I run animal experiments which involve sound recognition
and/ordiscrimination. Many various sound stimuli are used (e.g.,
tones, instruments, animal sounds (including birds), machines,
artficial sounds, knocks, bells etc.). Is it possible to describe
quantitatively such various sounds (along multiple dimensions, I
suppose) in order to find quantitative indices of dfference, either in
multidimensional space, or along a single dimension?
Any help would be greatly appreciated.
Department of Neurophysiology
Nencki Institute of Experimental Biology
3, Pasteur St., 02-093 Warsaw, Poland
tel. (48-22) 659 85 71 ex 379
fax (48-22) 822 53 42