Attention based temporal filtering of sensory signals for data redundancy reduction

Abstract

Since modern computational devices are required to store and process increasing amounts of data generated from various sources, efficient algorithms for identification of significant information in the data are becoming essential. Sensory recordings are one example where automatic and continuous storing and processing of large amounts of data is needed. Therefore, algorithms that can alleviate the computational load of the devices and reduce their storage requirements by removing uninformative data are important. In this work we propose a method for data reduction based on theories of human attention. The method detects temporally salient events based on the context in which they occur and retains only those sections of the input signal. The algorithm is tested as a pre-processing stage in a weakly supervised keyword learning experiment where it is shown to significantly improve the quality of the codebooks used in the pattern discovery process.

Modelling auditory attention

“…An integration of bottom-up and top-down modelling techniques replicating processes in the auditory pathway was demonstrated to improve sound localization in reverberant environments [84]. Auditory salience has been demonstrated to be an effective criterion for compression to reduce data size while retaining meaningful segments of large datasets of sound [85] and video [86]. Salience extraction has also been used as an abnormal sound detection mechanism for temporal signals, and generalized to lung sounds to use for finding medical abnormalities [87].…”

3PRO – An unsupervised method for the automatic detection of sentence prominence in speech

“…The most common approach in modeling perceptual attention is to look for unusual changes that take place in specific spatial or temporal contexts, therefore, covering methods that focus on looking for something rare, surprising, or novel (see, e.g., Itti & Baldi, 2009), looking for contrasts (see, e.g., Kakouros, Räsänen, & Laine, 2013), or maximizing the information gain from the input (see, e.g., Bruce & Tsotsos, 2009). Itti and Baldi (2009) have argued that surprisal exists only in the presence of uncertainty that can be described in a relative, subjective manner, based on the expectations of the observer (Itti & Baldi, 2009).…”