What is the neural representation of a speech code as it evolves in time? How do listeners integrate temporally distributed phonemic information across hundreds of milliseconds, even backwards in time, into coherent representations of syllables and words? What sorts of brain mechanisms encode the correct temporal order, despite such backwards effects, during speech perception? How does the brain extract rate- invariant properties of variable-rate speech? This article describes an emerging neural model that suggests answers to these questions, while quantitatively simulating challenging data about audition, speech and word recognition. This model includes bottom-up filtering, horizontal competitive, and top-down attentional interactions between a working memory for short-term storage of phonetic items and a list categorization network for grouping sequences of items. The conscious speech and word recognition code is suggested to be a resonant wave of activation across such a network, and a percept of silence is proposed to be a temporal discontinuity in the rate with which such a resonant wave evolves. Properties of these resonant waves can be traced to the brain mechanisms whereby auditory, speech, and language representations are learned in a stable way through time. Because resonances are proposed to control stable learning, the model is called an Adaptive Resonance Theory, or ART, model.

How does the brain extract invariant properties of variable-rate speech? A neural model, called PHONET, is developed to explain aspects of this process and, along the way, data about perceptual context effects. For example, in consonant vowel (CV) syllables such as /ba/ and /wa/, an increase in the duration of the vowel can cause a switch in the percept of the preceding consonant from /w/ to /b/ (Miller and Liberman, 1979). The frequency extent of the initial formant transitions of fixed duration also influences the percept (Schwab, Sawusch, and Nusbaum, 1981). PHONET quantitatively simulates over 98% of the variance in these data using a single set of parameters. The model also qualitatively explains many data about other perceptual context effects. In the model, C and V inputs are filtered by parallel auditory streams that respond preferentially to transient and sustained properties of the acoustic signal before being stored in parallel working memories. A lateral inhibitory network ...

Both animals and humans represent and compare numerical quantities, but only humans have evolved multi-digit place-value number systems. This article develops a Spatial Number Network, or SpaN, model to explain how these shared numerical capabilities are computed using a spatial representation of number quantities in the Where cortical processing stream, notably the inferior parietal cortex. Multi-digit numerical representations that obey a place-value principle are proposed to arise through learned interactions between categorical language representations in the What cortical processing stream and the Where spatial representation. Learned semantic categories that symbolize separate digits, as well as place markers like `ty,' `hundred,' and `thousand,' are associated through learning with the corresponding spatial locations of the Where representation. Such What-to-Where auditory-to-visual learning generates place-value numbers as an emergent property, and may be compared with other examples of multi-modal cross-modality learning, including synesthesia. The model quantitatively simulates error rates in quantification and numerical comparison tasks, and reaction times for number priming and numerical assessment and comparison tasks. In the Where cortical process, transient responses to inputs are integrated before they activate an ordered spatial map that selectively responds to the number of events in a sequence and exhibits Weber law properties. Numerical comparison arises from activity pattern changes across the spatial map that define a `directional comparison wave.' Variants of these model mechanisms have elsewhere been used to explain data about other Where stream phenomena, such as motion perception, spatial attention, and target tracking. The model is compared wi...

Temporal pattern classification normally operates upon vectors whose components are time-delayed samples, be they samples of spectral, cepstral, or LPC coefficients over time. Such a spatial representation of temporal patterns is very sensitive to speed variations, and requires computationally expensive time-alignment techniques such as Dynamic Time Warping in order to compare inputs to exemplars. This paper proposes a speed-invariant representation of temporal patterns using Taylor series expansion. A vector composed of successive time-derivative samples of the temporal signal is unique to the shape of the function in a region around the sampling point. Simple manipulations on such a "Taylor vector" yield a speed-invariant form. The degree of speed invariance can be controlled parametrically while retaining sensitivity to the direction of presentation. Simulations demonstrate that learning and recall of temporal pattterns coded using this representation is accurate and do...

normalization using cortical strip maps: