Modeling Two-Channel Speech Processing With the EPIC Cognitive Architecture.

An important application of cognitive architectures is to provide human performance models that capture psychological mechanisms in a form that can be "programmed" to predict task performance of human-machine system designs. Although many aspects of human performance have been successfully modeled in this approach, accounting for multitalker speech task performance is a novel problem. This article presents a model for performance in a two-talker task that incorporates concepts from psychoacoustics, in particular, masking effects and stream formation.

Enhancing listener strategies using a payoff matrix in speech-on-speech masking experiments.

Speech recognition was measured as a function of the target-to-masker ratio (TMR) with syntactically similar speech maskers. In the first experiment, listeners were instructed to report keywords from the target sentence. Data averaged across listeners showed a plateau in performance below 0 dB TMR when masker and target sentences were from the same talker. In this experiment, some listeners tended to report the target words at all TMRs in accordance with the instructions, while others reported keywords from the louder of the sentences, contrary to the instructions. In the second experiment, stimuli were the same as in the first experiment, but listeners were also instructed to avoid reporting the masker keywords, and a payoff matrix penalizing masker keywords and rewarding target keywords was used. In this experiment, listeners reduced the number of reported masker keywords, and increased the number of reported target keywords overall, and the average data showed a local minimum at 0 dB TMR with same-talker maskers. The best overall performance with a same-talker masker was obtained with a level difference of 9 dB, where listeners achieved near perfect performance when the target was louder, and at least 80% correct performance when the target was the quieter of the two sentences.

Perceiving sequential dependencies in auditory streams.

This study examined the ability of human listeners to detect the presence and judge the strength of a statistical dependency among the elements comprising sequences of sounds. The statistical dependency was imposed by specifying transition matrices that determined the likelihood of occurrence of the sound elements. Markov chains were constructed from these transition matrices having states that were pure tones/noise bursts that varied along the stimulus dimensions of frequency and/or interaural time difference. Listeners reliably detected the presence of a statistical dependency in sequences of sounds varying along these stimulus dimensions. Furthermore, listeners were able to discriminate the relative strength of the dependency in pairs of successive sound sequences. Random variation along an irrelevant stimulus dimension had small but significant adverse effects on performance. A much greater decrement in performance was found when the sound sequences were concurrent. Likelihood ratios were computed based on the transition matrices to specify Ideal Observer performance for the experimental conditions. Preliminary modeling efforts were made based on degradations of Ideal Observer performance intended to represent human observer limitations. This experimental approach appears to be useful for examining auditory "stream" formation and maintenance over time based on the predictability of the constituent sound elements.

State-space models of head-related transfer functions for virtual auditory scene synthesis.

This study investigates the use of reduced-order state-space models of collections of head-related transfer functions (HRTFs). Recent head-phone applications have motivated interest in binaural displays that can render multiple simultaneous virtual sound sources, acoustic reflections, and source and listener motion. In the present study, a multi-direction framework is considered that can render such phenomena by filtering source signals with a collection of HRTFs rather than individual HRTFs. The collection of HRTFs is implemented in the state-space, and approximation techniques are applied to construct low-order approximants that are indiscriminable from full-order HRTFs. Two experiments are described in which five observers are asked to discriminate between state-space and full-order renderings. Depending on the stimulus conditions and discrimination task, order thresholds of 7

Genetic algorithms for adaptive psychophysical procedures: recipient-directed design of speech-processor MAPs.

OBJECTIVES: The goal of the research is to evaluate the application of genetic algorithms (GAs) in listener-directed optimization of audio-processing designs. We hypothesize that cochlear-implant recipients can use a GA-guided adaptive psychophysical search procedure to select useful designs from among a large number of speech processor MAPS. DESIGN: An adaptive psychophysical procedure was developed in which a listener's preferred four out of eight speech processor MAPs were updated according to a genetic algorithm. Experiments involving cochlear-implant recipients were conducted to characterize both the convergence behavior of the adaptive procedure as well as properties of the MAPs optimized by the recipient. RESULTS: Results from five cochlear-implant recipients indicate that the adaptive procedure converges to useful speech processor MAPs within twenty iterations. CONCLUSION: The results suggest a means whereby a potentially large number of audio-processing designs can be searched efficiently by a human listener without requiring excessive amounts of feedback or prior knowledge about the listener's preferences. In the case of cochlear-implant recipients, it may be possible to use this procedure as an aid to the clinician in the fitting of a speech processor MAP.