We revisit the analysis by synthesis (A×S) approach to speech recognition. In the late 1950s and 1960s, Stevens and Halle proposed a model of spoken word recognition in which candidate word representations were synthesised from brief cues in the auditory signal and analysed against the input signal in tightly linked bottom-up/top-down fashion. While this approach failed to garner much support at the time, recent years have brought a surge of interest in Bayesian approaches to perception, and the idea of A×S has consequently gained attention, particularly in the domain of visual perception. We review the model and illustrate some data from speech perception that are well-accounted for in the context of such an architecture. We focus on prediction in speech perception, an operation at the centre of the A×S algorithm. The data reviewed here and the current possibilities to study online measures of speech processing using cognitive neuroscience methods, in our view, add to a provocative series of arguments why A×S should be reconsidered as a contender in speech recognition research, complementing currently more dominant models.

Are words stored as morphologically structured representations? If so, when during word recognition are morphological pieces accessed? Recent masked priming studies support models that assume early decomposition of (potentially) morphologically complex words. The electrophysiological evidence, however, is inconsistent. We combined masked morphological priming with magneto-encephalography (MEG), a technique particularly adept at indexing processes involved in lexical access. The latency of an MEG component peaking, on average, 220 msec post-onset of the target in left occipito-temporal brain regions was found to be sensitive to the morphological prime–target relationship under masked priming conditions in a visual lexical decision task. Shorter latencies for related than unrelated conditions were observed both for semantically transparent (cleaner–CLEAN) and opaque (corner–CORN) prime–target pairs, but not for prime–target pairs with only an orthographic relationship (brothel–BROTH). These effects are likely to reflect a prelexical level of processing where form-based representations of stems and affixes are represented and are in contrast to models positing no morphological structure in lexical representations. Moreover, we present data regarding the transitional probability from stem to affix in a post hoc comparison, which suggests that this factor may modulate early morphological decomposition, particularly for opaque words. The timing of a robust MEG component sensitive to the morphological relatedness of prime–target pairs can be used to further understand the neural substrates and the time course of lexical processing.

A long-standing question in speech perception research is how listeners extract linguistic content from a highly variable acoustic input. In the domain of vowel perception, formant ratios, or the calculation of relative bark differences between vowel formants, have been a sporadically proposed solution. We propose a novel formant ratio algorithm in which the first (F1) and second (F2) formants are compared against the third formant (F3). Results from two magnetoencephalographic experiments are presented that suggest auditory cortex is sensitive to formant ratios. Our findings also demonstrate that the perceptual system shows heightened sensitivity to formant ratios for tokens located in more crowded regions of the vowel space. Additionally, we present statistical evidence that this algorithm eliminates speaker-dependent variation based on age and gender from vowel productions. We conclude that these results present an impetus to reconsider formant ratios as a legitimate mechanistic component in the solution to the problem of speaker normalisation.

PurposeSpeech perception can be described as the transformation of continuous acoustic information into discrete memory representations. Therefore, research on neural representations of speech sounds is particularly important for a better understanding of this transformation. Speech perception models make specific assumptions regarding the representation of mid vowels (e.g., [ɛ]) that are articulated with a neutral position in regard to height. One hypothesis is that their representation is less specific than the representation of vowels with a more specific position (e.g., [æ]). MethodIn a magnetoencephalography study, we tested the underspecification of mid vowel in American English. Using a mismatch negativity (MMN) paradigm, mid and low lax vowels ([ɛ]/[æ]), and high and low lax vowels ([i]/[æ]), were opposed, and M100/N1 dipole source parameters as well as MMN latency and amplitude were examined. ResultsLarger MMNs occurred when the mid vowel [ɛ] was a deviant to the standard [æ], a result consistent with less specific representations for mid vowels. MMNs of equal magnitude were elicited in the high–low comparison, consistent with more specific representations for both high and low vowels. M100 dipole locations support early vowel categorization on the basis of linguistically relevant acoustic–phonetic features. ConclusionWe take our results to reflect an abstract long-term representation of vowels that do not include redundant specifications at very early stages of processing the speech signal. Moreover, the dipole locations indicate extraction of distinctive features and their mapping onto representationally faithful cortical locations (i.e., a feature map).