Kenneth N. Stevens - US grants

Develop and evaluate an electroaccoustical device which will reliably and validly measure hearing thresholds in the frequency range of 8000-20000 Hz. The ultimate aim of this work will be a system capable of assessing the extent of high-frequency hearing impairment in humans.

The long range goal of this research is to develop models of the process whereby a speaker encodes a discrete linguistic representation of an utterance into patterns of movement of the articulatory structures and then into patterns of sound. Progress toward the development of these models makes it possible to contribute toward a second more practical goal. This second goal is to devise procedures for aiding with the assessment of deviations from normal functioning of the speech generation process through measurements of the acoustic pattern that results from this disordered speech production. The models can then be used to infer the movements of the articulators from the acoustic measurements. The development of models of the normal speech production process will be advanced through a combination of theoretical analysis of sound generation and filtering in the vocal tract, and collection of acoustic data from a variety of utterances of normal speakers. The acoustic data will be augmented in some cases by physiological data and by perceptual studies with synthetic speech. The acoustic data will be used for two purposes. One is to provide support for the acoustic theory that attempts to predict sound patterns when articulatory movements are known or can be estimated. The second is to provide some basic information about the ways in which speakers manipulate and coordinate different articulators when particular sequences of linguistic units (syllables, segments, features) are encoded into sound. In the proposed research, emphasis will be placed on the production of liquids, glides, nasal consonants, nasal vowels, and fricative and stop consonants. Acoustic data will be used to infer the movements of the articulators for segment sequences consisting of certain consonant clusters within and across syllable boundaries and to examine laryngeal adjustments in sentences in syllables having different degrees of prominence or reduction. The development of procedures for using acoustic measurements to assist in evaluating speech disorders will involve collaboration with clinicians. A limited number of disorders will be studied, and samples of speech from individuals with these disorders will be analyzed exhaustively. The results of the analysis will be interpreted to infer deviations in the control of movements of the different articulatory structures. The gains to be achieved by augmenting conventional assessment methods with acoustic analysis will be addressed.

The aim of the proposed work is to obtain experimental evidence that bears on three topics: (1) the features that characterize the segments of language, (2) the surface form that segments take when they are concatenated into larger units, and (3) models of the processing by which concatenation occurs. In addition to the experimental studies, we will develop theories and models that attempt to account for the influence of characteristics of the human perception, production and cognition systems on all three of these areas. Research concerned with topic (1) will involve collecting data on the production, acoustic form and perception of various classes of speech sounds, together with experimental studies and modelling of some basic properties of the human sound-generating and auditory perception systems. Research directed toward topic (2) will include studies of the types of phonetic variation that occur in various languages, particularly English, and how this variation interacts with linguistic factors and with biomechanical factors and principles of speech motor control. Research on the speech planning process (topic 3) includes studies of coarticulation planning, and determination of acoustic-phonetic rules and their ordering in fluent speech. Other experiments will examine speech errors (using an error-elicitation paradigm) and explore segment-manipulating abilities of speakers (involving "play-languages" like Pig Latin). Results will be interpreted in terms of models for speech planning.

The aim of the proposed work is to obtain experimental evidence that bears on three topics: (1) the features that characterize the segments of language, (2) the surface form that segments take when they are concatenated into larger units, and (3) models of the processing by which concatenation occurs. In addition to the experimental studies, we will develop theories and models that attempt to account for the influence of characteristics of the human perception, production and cognition systems on all three of these areas. Research concerned with topic (1) will involve collecting data on the production, acoustic form and perception of various classes of speech sounds, together with experimental studies and modelling of some basic properties of the human sound-generating and auditory perception systems. Research directed toward topic (2) will include studies of the types of phonetic variation that occur in various languages, particularly English, and how this variation interacts with linguistic factors and with biomechanical factors and principles of speech motor control. Research on the speech planning process (topic 3) includes studies of coarticulation planning, and determination of acoustic-phonetic rules and their ordering in fluent speech. Other experiments will examine speech errors (using an error-elicitation paradigm) and explore segment-manipulating abilities of speakers (involving "play-languages" like Pig Latin). Results will be interpreted in terms of models for speech planning.

This three-year continuing grant will support a study of word recognition in continuous speech. New principles of phonetic and acoustic indexing will be developed through analysis of selected utterances gathered under controlled conditions. Landmark features in acoustic signals will be identified and characterized, together with quantifiable relationships of other features to these landmarks. The investigators will produce a small lexicon indexed by these phonetic features, test its effectiveness in word recognition, and begin development of a system for automatic extraction of phonetic features. The ultimate impact may be a new theory of phonetic encoding and a more efficient method of automatic speech recognition.

This two-year award in linguistics will support collaborative research between Dr. Kenneth N. Stevens of the Massachusetts Institute of Technology (M.I.T) and Dr. Gunnar Fant of the Royal Institute of Technology (R.I.T) in Stockholm, Sweden. The objective of the proposed project is to advance scientific understanding of the acoustic processes of speech sounds, specifically consonant production. The M.I.T and R.I.T speech groups have had a long-standing collaboration in theoretical studies of acoustics and perception of consonants. Recently, the R.I.T group has been active in the development of computer-based vocal tract-models of consonant production; whereas, the MIT group has focused on acoustic analysis and synthesis of consonants and the application of acoustic studies to theories of phonetics and phonology. The proposed collaboration will link the recent work of American and Swedish speech groups in problematic areas in consonant production. These include: (1) sound generation at the larynx at vowel- consonant and consonant-vowel boundaries, (2) mechanisms of sound production at the release of stop consonants; and (3) acoustic changes at the implosion and release of nasal consonants. Dr. Stevens proposes to use the computer-based model at R.I.T to extensively test the M.I.T theories and to understand the model so it can be implemented on computers at M.I.T. for future research work. The field will benefit from the complementary expertise of both groups and advance our understanding of consonant production and processes. This research has important applications in the development of machine-made (electronic) speech synthesis and recognition. It also has relevance for the diagnosis and treatment of communication disorders and in second language learning.

The aim of the proposed work is to obtain experimental evidence that bears on three topics: (1) the features that characterize the segments of language, (2) the surface form that segments take when they are concatenated into larger units, and (3) models of the processing by which concatenation occurs. In addition to the experimental studies, we will develop theories and models that attempt to account for the influence of characteristics of the human perception, production and cognition systems on all three of these areas. Research concerned with topic (1) will involve collecting data on the production, acoustic form and perception of various classes of speech sounds, together with experimental studies and modelling of some basic properties of the human sound-generating and auditory perception systems. Research directed toward topic (2) will include studies of the types of phonetic variation that occur in various languages, particularly English, and how this variation interacts with linguistic factors and with biomechanical factors and principles of speech motor control. Research on the speech planning process (topic 3) includes studies of coarticulation planning, and determination of acoustic-phonetic rules and their ordering in fluent speech. Other experiments will examine speech errors (using an error-elicitation paradigm) and explore segment-manipulating abilities of speakers (involving "play-languages" like Pig Latin). Results will be interpreted in terms of models for speech planning.

The aim of the proposed work is to obtain experimental evidence that bears on three topics: (1) the features that characterize the segments of language, (2) the surface form that segments take when they are concatenated into larger units, and (3) models of the processing by which concatenation occurs. In addition to the experimental studies, we will develop theories and models that attempt to account for the influence of characteristics of the human perception, production and cognition systems on all three of these areas. Research concerned with topic (1) will involve collecting data on the production, acoustic form and perception of various classes of speech sounds, together with experimental studies and modelling of some basic properties of the human sound-generating and auditory perception systems. Research directed toward topic (2) will include studies of the types of phonetic variation that occur in various languages, particularly English, and how this variation interacts with linguistic factors and with biomechanical factors and principles of speech motor control. Research on the speech planning process (topic 3) includes studies of coarticulation planning, and determination of acoustic-phonetic rules and their ordering in fluent speech. Other experiments will examine speech errors (using an error-elicitation paradigm) and explore segment-manipulating abilities of speakers (involving "play-languages" like Pig Latin). Results will be interpreted in terms of models for speech planning.

speech recognition; model design /development; speech; cues; language development; behavioral /social science research tag; human subject; clinical research;

The long-range goal of this project is to develop a model that simulates the human perception of the words in spoken language. The immediate goal in the present project is to achieve one part of this task: the identification of the distinctive features of consonants in spoken syllables, words and sentences. Key ideas in our approach are that the speech signal contains reliable acoustic cues to the articulation implemented by the speaker even when surface phonetics are variable, and that the articulation is governed by abstract contrastive phonological representations. Identification of the consonant features proceeds by first performing an acoustic analysis of the speech to establish the locations of landmarks or discontinuities in the sound where consonant closures and releases are formed. The sound in the vicinity of those landmarks is then subjected to further detailed analysis to establish the underlying features of the consonant that generates each landmark, including the place of articulation, the voicing feature, the nasal feature etc. This further analysis involves extracting from the sound a number of attributes that provide cues for each of the underlying consonant features. The selection of these attributes is guided by the requirement that they be closely related to the articulatory shapes and movements that produced the speech. In the proposed work, our current understanding of the combination of attributes that most effectively reveal the articulation and its governing features and segments will be expanded and refined, through detailed theory-driven acoustic measures, perceptual experimentation and appropriate statistical analysis. The robustness of the model will be evaluated using various kinds of utterances, from citation forms to running speech. The performance of the model will also be tested in speech that has been contaminated with noise, and the errors made by the model will be compared with those made by human listeners. The model has application in the study of speech perception by listeners with impaired hearing or by listeners in an environment in which speech is degraded. Understanding of these processes of speech perception can lead to improved approaches to the remediation of disorders of speech perception and production.