A growing number of affective computing researches recently developed a computer system that can recognize an emotional state of the human user to establish affective human-computer interactions. Various measures have been used to estimate emotional states, including self-report, startle response, behavioral response, autonomic measurement, and neurophysiologic measurement. Among them, inferring emotional states from electroencephalography (EEG) has received considerable attention as EEG could directly reflect emotional states with relatively low costs and simplicity. Yet, EEG-based emotional state estimation requires well-designed computational methods to extract information from complex and noisy multichannel EEG data. In this paper, we review the computational methods that have been developed to deduct EEG indices of emotion, to extract emotion-related features, or to classify EEG signals into one of many emotional states. We also propose using sequential Bayesian inference to estimate the continuous emotional state in real time. We present current challenges for building an EEG-based emotion recognition system and suggest some future directions.

Neff and Green [Percept. Psychophys. 41, 409-415 (1987)] report that the masking of single tones by random-frequency multitone maskers varies nonmonotonically with number of masker components (peaking at 10-50 components). In this paper it is shown that such results are well predicted by a model (the component-relative-entropy model, CoRE) wherein thresholds increase linearly with the ensemble variance of masker spectra smoothed by peripheral auditory filters [R. A. Lutfi, J. Acoust. Soc. Am. 94, 748-758 (1993)]. Three experiments were conducted. In the first, the nonmonotonic relation was replicated for 9 of 11 listeners in conditions similar to those of Neff and Green. In the second, the frequencies of masker components were fixed and the levels of components were varied randomly across presentations to simulate Gaussian noise. In this case, the nonmonotonicity and the total amount of masking for these listeners were shown to be significantly reduced. In the third experiment, masked thresholds for the signal were found to vary monotonically with the frequency spacing of masker components for a fixed number of masker components. Large individual differences among listeners were obtained in some experimental conditions. Individual as well as mean thresholds were well predicted by the CoRE model with an appropriate selection of the values of the two free parameters of the model for each listener.

Preschoolers and adults were asked to detect a 1000-Hz signal, which was masked by a multitone complex. The frequencies and amplitudes of the components in the complex varied randomly and independently on each presentation. A staircase, cued two-interval, forced-choice procedure disguised as a "listening game" was used to obtain signal thresholds in quiet and in the presence of the multitone maskers. The number of components in the masker was fixed within an experimental condition and varied from 2 to 906 across experimental conditions. Thresholds were also measured with a broadband noise masker. Eight preschool children and eight adults were tested. Although individual differences were large, among both adults and children, there was little difference between the groups in the mean amount of masking produced by the maskers with large numbers of components (400 and 906). There was also a small but significant difference between adults and children in the mean amount of masking produced by the broadband noise. The difference between the groups was much larger with smaller numbers of components. Data obtained from the adults were basically similar to that previously reported [cf. Neff and Green, Percept. Psychophys. 41, 409-415 (1987); Oh and Lutfi, J. Acoust. Soc. Am. 104, 3489-3499 (1998)]: maskers comprised of 10-40 components produced as much as 30 to 60 dB of masking in some, but not all listeners. Those same maskers produced larger amounts of masking (70-83 dB) in many of the preschool children, although, as in the adult group, individual differences were large. The component-relative-entropy (CoRE) model [Lutfi, J. Acoust. Soc. Am. 94, 748-758 (1993)] was used to describe the differences in performance between the children and adults. According to this model the average child appears to integrate information over a larger number of auditory filters than the average adult.

When normal-hearing adults and children are required to detect a 1000-Hz tone in a random-frequency multitone masker, masking is often observed in excess of that predicted by traditional auditory filter models. The excess masking is called informational masking. Though individual differences in the effect are large, the amount of informational masking is typically much greater in young children than in adults [Oh et al., J. Acoust. Soc. Am. 109, 2888-2895 (2001)]. One factor that reduces informational masking in adults is spatial separation of the target tone and masker. The present study was undertaken to determine whether or not a similar effect of spatial separation is observed in children. An extreme case of spatial separation was used in which the target tone was presented to one ear and the random multitone masker to the other ear. This condition resulted in nearly complete elimination of masking in adults. In young children, however, presenting the masker to the nontarget ear typically produced only a slight decrease in overall masking and no change in informational masking. The results for children are interpreted in terms of a model that gives equal weight to the auditory filter outputs from each ear.

Masked threshold for a pure-tone signal can be substantially elevated whenever the listener is uncertain about the spectral or temporal properties of the masker, an effect referred to as auditory informational masking. Individual differences in the effect are large, with young children being most susceptible. When masker uncertainty is introduced by randomizing the frequencies of a multitone masker on each presentation, the function relating a child's pure-tone signal threshold to the number of masker components is found to be substantially elevated above that of most adults. The age effect and the individual differences among adults are not well understood, though a difference in the shapes of the masking functions suggests that different detection strategies may be involved. The present study reports results from a principal components analysis of informational masking functions obtained from 38 normal-hearing children ranging in age from 4 to 16 years and 46 normal-hearing adults ranging in age from 19 to 38 years. The premise underlying the analysis is that if different detection strategies are involved, they should add independent sources of variance to the masking functions. Hence, more than one principal component (PC) should be required to account for a substantial proportion of the variance in these functions. The results, instead, supported the operation of a single underlying strategy with all but 17% of the variance accounted for by the first PC within and across age groups. An analysis of variance on the first two PCs showed that only the first changed with age, and a cluster analysis of the masking functions showed complete separation of clusters along this PC for all but 1 listener. The results are taken to suggest that large individual differences in informational masking at all ages reflect differences in the extent to which masker uncertainty adds variance to the decision variable of an otherwise optimal decision strategy.