Abstract not found

Physiological studies of speech production have demonstrated that even simple articulation involves a range of specialized motor and cognitive processes and the neural mechanisms responsible for speech re¯ect this complexity. Recently, a number of functional imaging techniques have contributed to our knowledge of the neuroanatomical and neurophysiological correlates of speech production. These new imaging approaches have the advantage of permitting study of large numbers of normal and disordered subjects but they bring with them a host of new methodological concerns. One of the challenges for understanding language production is the recording of articulation itself. The problems associated with measuring the vocal tract and measuring the neural activity during overt speech are reviewed. It is argued that advances in understanding fundamental questions such as what are the planning units of speech, what is the role of feedback during speech and what is the in¯uence of learning, await the development of better methods for assessing task performance. Ó 2001 Elsevier Science B.V. All rights reserved.

A series of arguments is presented showing that words are not stored in memory in a way that resembles the abstract, phonological code used by alphabetical orthographies or by linguistic analysis. Words are stored in a very concrete, detailed auditory code that includes nonlinguistic information including speaker’s voice properties and other details. Thus, memory for language resembles an exemplar memory and abstract descriptions (using letter-like units and speaker-invariant features) are probably computed on the fly whenever needed. One consequence of this hypothesis is that the study of phonology should be the study of generalizations across the speech of a community and that such a description will employ units (segments, syllable types, prosodic patterns, etc.) that are not necessarily employed as units in speakers’ memory for language. That is, the psychological units of language are not useful for description of linguistic generalizations and linguistic generalizations across a community are not useful for storing the language for speaker use.

This chapter describes how an artificial device, able to produce acoustic signals from articulatory motion, can learn to speak, i.e. coordinate its articulatory movements in such a way that it utters meaningful sequences of sounds belonging to a given language. This complex learning procedure, accomplished within a few years by the human child, is simulated in four major steps: (a) a babbling phase, where the device builds up a model of the forward kinematics, i.e. the articulatory-to-audio-visual mapping; (b) an imitation stage, where it tries to reproduce a limited set of sound sequences by audio-visual-to-articulatory inversion including a normalization procedure; (c) a "shaping" stage, where phonemes are associated with sensori-motor representation; and finally, (d) a "rhythmic" phase, where it learns the appropriate coordination of the activations of these sensori-motor targets. This artificial device has thus an ear which delivers both the control signals and the identification of pe...

Phonological grammars try to represent speakers ’ knowledge so that the ‘natural’ behavior of speech sounds becomes self-evident. Phonetic models have the same goals but have no psychological pretensions. Phonetic models succeed in explaining the natural behavior of speech, whereas phonological representations largely fail. The ‘phonetic naturalness ’ requirement in phonological grammars should be re-examined and probably abandoned.

The well known contrast in Russian between palatalized and non-palatalized consonants originated roughly one thousand years ago. At that time consonants were allophonically palatalized before front vowels, as in dan j Z 'tribute'. When the 'jer' (high, lax) vowels disappeared in certain positions, the palatalization formerly triggered by the front jer remained, leading to a palatalization contrast across most consonant types, e.g., dan j 'tribute' vs. dan 'given' (< dan). At the same time or soon thereafter, a rule is said to have been established by which /i/ surfaced as [Y] after non-palatalized consonants, e.g., ot Ym en i 'on behalf of' (< ot im en i). This paper analyzes these two sound changes within a version of Dispersion Theory (DT, Flemming 1995a) elaborated by N Chiosin and Padgett (to appear) and Padgett (1997, 2001a). DT differs from other current models of phonology in its fundamentally systemic orientation: constraints evaluate not only isolated forms as is usual, but sets of forms in contrast. References to these systems of contrast is key to the statement of constraints governing the perceptual distinctiveness of contrasts on the one hand, and constraints directly penalizing merger (neutralization) on the other. The analysis of the Russian facts here illustrates how this theory works, and provides an explanation for the emergence of the /i/ # [Y] rule upon the fall of the jers.

In the last 50 years there have been steady gains in phonetic knowledge and punctuated equilibrium in phononological theories. Phonetics and phonology meet most obviously in the definition of the set of features used to describe phonological processes. The Jakobsonian statement of distinctive feature theory in the 1952 caused a paradigm shift in the relations between phonetics and phonology. Changes occurred again with the introduction of a mentalist view of phonological features in the 1968 publication of The Sound Pattern of English. In1972 Stevens ’ introduced Quantal theory and the hunt for acoustic invariance for phonological features was on. Autosegmental phonology and notions of a feature hierarchy brought further demands on phonetic knowledge. Now Optimality theory has proposed a new way of relating phonological contrasts and phonetic data,and Articulatory phonology has spurred great phonetic progress that is just beginning to have a direct impact on phonology.. FEATURAL BEGINNINGS It’s nice to be asked to present a historical survey of half a century of phonetics and phonology in this MIT meeting, because there was a clear cut change in the relations between these fields

Voice quality conveys both linguistic and paralinguistic information, and can be distinguished by acoustic source characteristics. We label objective voice quality categories based on the spectral and temporal structure of speech sounds, specifically the harmonic structure (H1-H2) and the mean autocorrelation ratio of each phone. Results from a classification experiment using a Support Vector Machine (SVM) show that allophones that differ from each other regarding voice quality can be classified using input features in speech recognition. Among different possible ways to incorporate voice quality information in speech recognition, we demonstrate that by explicitly modeling voice quality variance in the acoustic models using hidden Markov models, we can improve word recognition accuracy. Keywords: ASR, Voice quality, H1-H2, Autocorrelation ratio, SVM, HMM. 1

Abstract. Voice quality has been defined as the characteristic auditory colouring of an individual's voice, derived from a variety of laryngeal and supralaryngeal features and running continuously through the individual's speech. The distinctive tone of speech sounds produced by a particular person yields a particular voice. Voice quality is at the centre of several speech processing issues. In speech recognition, voice differences, particularly extreme divergences from the norm, are responsible for known performance degradations. In speech synthesis on the other hand, voice quality is a desirable modelling parameter, with millions of voice types that can be distinguished theoretically. This article reviews the experimental derivation of voice quality markers. Specifically, the use of perceptual judgements, the long-term averaged spectrum (LTAS) and prosodic markers is examined, as well as inverse filtering for the extraction of the glottal source waveform. This review suggests that voice quality is best investigated as a multi-dimensional parameter space involving a combination of factors involving individual prosody, temporally structured speech characteristics, spectral divergence and voice source features, and that it could profitably complement simple linguistic prosodic model processing in speech synthesis. 1

In this paper the issue of the nature of the representations of the speech production task in the speaker's brain is addressed in a production-perception interaction framework. Since speech is produced to be perceived, it is hypothesized that its production is associated for the speaker with the generation of specific physical characteristics that are for the listeners the objects of speech perception. Hence, in the first part of the paper, four reference theories of speech perception are presented, in order to guide and to constrain the search for possible correlates of the speech production task in the physical space: the Acoustic Invariance Theory, the Adaptive Variability Theory, the Motor Theory and the Direct-Realist Theory. Possible interpretations of these theories in terms of representations of the speech production task are proposed and analyzed. In a second part, a few selected experimental studies are presented, which shed some light on this issue. In the conclusion, on the basis of the joint analysis of theoretical and experimental aspects presented in the paper, it is proposed that representations of the speech production task are multimodal, and that a hierarchy exists among the different modalities, the acoustic modality having the highest level of priority. It is also suggested that these representations are not associated with invariant characteristics, but with regions of the acoustic, orosensory and motor control spaces. 1.