Naslov
A novel auditory saliency prediction model based on spectrotemporal modulations

Autori
Filipan, Karlo ; Bockstael, Annelies ; De Coensel, Bert ; Schönwiesner, Marc ; Botteldooren, Dick

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

ISBN
978-987-24713-6-1

Skup
22nd International Congress on Acoustics (ICA 2016)

Mjesto i datum
Buenos Aires, Argentina, 05.09.2016. - 09.09.2016

Vrsta sudjelovanja
Pozvano predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
attention ; saliency ; modeling ; ripples

Sažetak
Previous studies indicate that soundscape perception and appraisal are influenced by the soundsthat people hear and pay attention to. Hence, a model that evaluates instantaneous humanattention to environmental sounds would be very useful in soundscape research. Attention istriggered by the saliency of a sound within its context. Therefore, we propose a model for predict-ing saliency of sounds based on dynamic modulation ripples – simultaneous modulations in thefrequency and time domain. These ripples exhibit direct response in the auditory cortex of thehuman brain. Our model contains three stages. In the first stage, the incoming sound signal isdemodulated similarly to the early stages of auditory processing, and afterwards it is correlatedwith each of the modulation functions of the ripples. The obtained ripple features enable themodel to detect salient changes that are not accompanied by changes in more commonly usedspectrogram features. We demonstrate this by comparing the model output for sound signals withthe same amplitude but randomized phase spectrum. The second stage of the model integratesripple features over time to simulate excitation and inhibition processes happening along neuralpathways. In the final stage, spectral saliency is aggregated to an overall saliency using super-vised training on sound environments with embedded salient sounds. We evaluate the modelwith a collection of natural sound fragments previously used in an EEG experiment on attentionand illustrate its application in complex environmental sound scenes.

Izvorni jezik
Engleski

Znanstvena područja
Elektrotehnika, Računarstvo

POVEZANOST RADA