Phonetic Targets Research Articles

Peak alignment in accented Tolowa syllables

Tolowa (Athabaskan, Pacific Coast Group) has only three remaining fluent speakers near Crescent City, CA. Tolowa has prominent pitch contours that have been described both as an interaction of stress and vowel length and as lexically specified pitch accent. The present study examined the phonetic realization of these pitch contours. Fundamental frequency (F0) at its highest peak during the rhyme of accented syllables was measured, as was relative duration into the rhyme at which the peak occurred. Preliminary results indicated that a high pitch peak occurred approximately 20%–30% of the way into the rhyme of the syllable. In the case of long vowels in accented syllables, this led to a prominent peak which fell sharply during the second half of the rhyme. A similar peak and fall was seen in accented syllables closed by a voiced sonorant. In the case of accented syllables with short vowels closed by a voiceless consonant, the accent was realized as a rise in pitch leading to the final consonant. These results indicate that the phonetic target is a high peak 20%–30% into the rhyme; the target seems to not include the fall seen in rhymes with long vowels or those closed by voiced sonorants.

The PROMPT model: Use and application for children with mixed phonological-motor impairment

PROMPT (Prompts for Restructuring Oral Muscular Phonetic Targets) a tactually grounded, sensori-motor, cognitivelinguistic treatment model is explored as a philosophy, approach, system and technique. The PROMPT Conceptual Framework and the Motor-Speech Hierarchy are described as frameworks for both assessment and treatment that help clinicians develop a holistic communication focus for treatment, while embedding motor, language and social interaction goals. The role of technique and tactual systems in PROMPT are explained with relevance to their use and application with children who have moderate to severe mixed phonological and motor impairment. General assessment guidelines and priorities for treatment, with one case study are discussed.

Exploring the ‘‘hyperspace’’ effect in vowel perception

When using the method of adjustment to choose the best representative of an English vowel in an F1 by F2 space, listeners have been found to select formant values more extreme than those they actually produce [Johnson et al., Language 69, 505–528 (1993)]. This preference for ‘‘hyperspace’’ targets (those beyond actual productions) is argued to indicate a phonetic target more extreme than the subjects’ actual hyperarticulations. The same methodology was used with 10 Rhode Island subjects. They produced vowels under two conditions, citation and then (after the perceptual task) hyperarticulated speech. Perception was assessed with a grid of 298 F1/F2 combinations from which listeners chose the best match for a keyword’s vowel. Females and males were nearly equivalent in perception, as assumed previously but not shown. However, only one vowel was clearly extreme in both formants, with seven other vowels being more extreme on one formant, two vowels showing hypoarticulation, and the final one showing no effect. This lack of a robust effect may be due to difficulty of using F2 in perceiving the synthetic vowels. It may also be that the listeners were selecting perceptual targets from a more neutral dialetic than their own. [Work supported by NIH.]

Phonetic targets as the link between speech production and speech perception

We argue in this paper that the coherence between speech perception and production, which the motor theory maintains, lies in the common knowledge about speech production shared by the speaker and the listener. More specifically, we propose that abstract phonemic units are associated with phonetic targets that are articulatorily operable. These targets, however, are usually not fully achieved in speech production. Rather, they are often only approximated by different degrees under the limitation of various articulatory constraints. The process of speech production, therefore, is one in which various phonetic targets are being continuously approximated. As evidence, we will demonstrate with recent acoustic data how lexical tones in Mandarin are implemented through continuous approximation under specific articulatory constraints, how these constraints bring about extensive variations in the F0 contours, and how, despite the variations, tonal targets maintain their integrity in connected speech. We will then further argue that listeners, being also speakers themselves, should be able to discover the intended targets by noting the consistency the speaker manages to achieve despite the articulatory constraints. In other words, listeners hear not just what the speakers do, but more importantly, what they intend to do: We listen to hear what is being said.

Sex, Sound Symbolism, and Sociolinguistics

Two important general conclusions of sociolinguistics are that females tend to lead in linguistic changes and that vowel systems tend to rotate in fixed directions. We argue that these two results are linked in that females and males are attracted asymmetrically to different subjective poles of the vowel system, one being high front unrounded [i] and the other being somewhere in the back rounded area near {o o}. We therefore adapt Ohala's “frequency code” theory to the sociolinguistic study of phonetic variation. We argue that such a model is superior to one in which individual speakers are attracted to particular phonetic targets only by virtue of a logically prior attraction to a class or similar social category which happens to use the pronunciation in question. Our study is in line with more general efforts to construct causal models for correlations between variable aesthetic preferences and “hard” (nonsubjective) social categories such as sex. It is, however, an open question whether sex‐asymmetric preferences are wired‐in or by‐products of more general developmental phenomena sensitive to gender socialization patterns.

Long-term effectiveness of PROMPT treatment in a severely apractic-aphasic speaker

This study examined the acquisition and long-term maintenance of a functional core vocabulary by a severely apractic-aphasic speaker following the application of Prompts for Restructuring Oral Muscular Phonetic Targets (PROMPT) treatment. The subject was a 24-year-old male who had suffered a single left-hemisphere thrombotic CVA approximately 2 years prior to the beginning of this investigation. Treatment and maintenance were monitored over a 41-week period. The results showed that the 30 target words and phrases were produced accurately during the treatment phases of the study and after treatment was discontinued.

Coarticulation modeling by linear filters in a pseudo-articulatory domain

A speech synthesizer for use in analysis-by-synthesis speech recognition is described. Coarticulation is modeled by applying linear FIR filters to target gestures in a pseudo-articulatory domain. This domain is treated as a hidden, unobservable layer between the phonetic input to the synthesizer and the acoustic spectrum output. That output is obtained from the hidden-layer signal by means of a memoryless nonlinear transformation that is implemented by a neural net with elliptic basis functions. The entire model, including phonetic targets, FIR filter shapes, and neural-net parameters, is trained by a pre-conditioned conjugate gradient method, using the mean-squared error between synthetic and actual spectra as the objective function. The gradient is calculated by a back-propagation algorithm. It is found that after training, the FIR filter shapes typically resemble noncausal two-pole lowpass characteristics, with one pole in the right and the other in the left half-plane, representing right-to-left (anticipatory) and left-to-right coarticulation, respectively. Performance of the synthesizer on isolated-word speech, continuous read speech, and spontaneous speech will be described, and results from recognition experiments on speaker-dependent and speaker-independent tasks will be reported.

The Hyperspace Effect: Phonetic Targets Are Hyperarticulated

Il est communement suppose, mais raremant justifie, que le processus de la reduction phonetique s'applique aux cibles phonetiques hyperarticulees. Le resultat de cette etude experimentale valide l'hypothese que les cibles phonetiques sont hyperarticulees et conformes a un modele de la realisation phonetique a deux phases : l'accouplement des traits distinctifs avec les cibles phonetiques hyperarticulees au niveau de la premiere phase et la reduction phonetique de ces cibles au niveau de la seconde

Differential Treatment of Phonological Disability in Children With Impaired Hearing Who Were Trained Orally

The efficacy of differential treatment of phonological disability was studied in children with moderate to severe sensorineural hearing impairments who presented with both phonetic and phonemic error types. Two treatments were administered to four subjects aged 5:0 to 10:5 with prelinguistic hearing impairment who had been trained orally. Phonetic treatment was modeled in accord with an articulation approach using syllable imitation. Phonemic treatment was modeled in accord with a phonological approach using meaningful minimal contrasts. Phonetic targets were consonant singletons; phonemic targets were phonological processes. Within each treatment, one target was trained; the other served as a control. Results indicated a tendency toward improved target production in training and generalization with phonemic treatment. Individual subject differences in training and in generalization were evident for the phonetic treatment. No between-treatment differences in training were found. However, between-treatment differences in generalization were significant. Clinical issues and implications of the findings are discussed.

An Investigation of Attention Allocation Deficits in Aphasia

Nine aphasic and 18 normal adults participated in a dual task with components of phoneme monitoring and semantic judgment. Subjects listened to lists of recorded spoken words and were required to detect semantic and phonetic targets. Two within-subject variables were manipulated: the probability of target occurrence for each dual-task component (.2, .5, and .8) and the explicitness of instruction (explicit vs. implicit). Only in the explicit condition were subjects told about the probability structure and given the attention allocation strategy congruent with that structure. Reaction times were recorded on-line for analysis by an IBM-AT. The results showed that normal subjects′ overall latencies in detecting phonetic targets decreased as the target occurrence probability increased in both implicit and explicit instruction conditions. The detection latencies for semantic targets showed a similar probability effect but only in the explicit condition. By contrast, the aphasic subjects did not show the probability effect in any condition, although they performed above the chance level on both semantic and phonetic tasks. The presence of the probability effects for the normal subjects indicated that an efficient attention allocation mechanism was operating so as to optimize the performance level. On the other hand, the absence of probability effects for the aphasic subjects suggests that their attention control system might be inefficient so that task-demand cues (probability and attention allocation instruction) were not utilized properly.

Temporal decomposition of speech

In articulatory phonetics speech is described as a sequence of distinct articulatory gestures, each of which produces an acoustic event that should approximate a phonetic target. Due to the overlap of the gestures these phonetic targets are often only partly realized.Atal (1983) has proposed a method for speech coding based on so-called temporal decomposition of speech into a sequence of overlapping target functions and corresponding target vectors. The target vectors may be associated with ideal articulatory positions. The target functions describe the temporal evolution of these targets. This method makes no use of specific articulatory or phonetic knowledge. We have extended and modified this method to improve the determination of the number and the location of the target functions and to overcome the shortcomings of the original method. With these improvements temporal decomposition has become a strong tool in analysing speech, from which researchers working on speech coding, recognition and synthesis may profit.

Automatic target generation for vowels

Research in phonetic recognition by computer is of interest because it may offer better performance than systems using larger‐recognition units such as syllables or words. Among several techniques developed to model speech signals as phoneme sequences (e.g., hidden Markov models and other feature‐based methods) is the phonetic‐target model based on the auditory‐perceptual theory of Miller [ASHA Reports 14 (1984)]. In it: (1) talker differences that mask acoustic phonetic correlates are minimized when incoming speech signals are represented as paths in a three‐dimensional space, where X = log(P3/P2), Y = log(P1/R), and Z = log(P2/P1); and (2) phonetic targets defined in this space seem to be uniquely related to the acoustic patterns of phonelike elements. An automatic procedure generated three phonetic targets for the vowels /ɪ/, /ɛ/, and /ʌ/ from a set of training tokens in stop‐vowel‐stop format recorded from two male and two female talkers. When four new talkers were tested, a vowel recognition accuracy of about 98 % was attained. Additionally, a segmentation algorithm isolated the vocalic segments from the stop bursts with 100% accuracy. [NINCDS grant R01‐NS21994‐01.]

Synthesis by rule of Japanese

Based on an analysis of a corpus of recorded CVCV syllables and several paragraphs [C. Aoki, D. Klatt, and H. Kawasaki, “Acoustic‐Phonetic Analysis of Japanese,” J. Acoust. Soc. Am. Suppl. 1 75, S60 (1984)], as well as a study of the prior literature on analysis and synthesis of Japanese, we have formulated a set of synthesis rules within the general framework used in DECtalk. Input to the system must be specified phonemically. The program is divided into several subprograms that (1) parse this string into phoneroes and associated structural/accent features, (2) apply phonological rules to select appropriate allophones or delete segments, (3) assign a duration to each segment, (4) specify onset times and strength of pitch rises and falls, (5) compute 17 time functions to control a formant synthesizer on the basis of stored tables of phonetic target values and smoothing time constants, and, finally, (6) compute a waveform from the control parameter specification. The program has been optimized by systematic spectral/waveform comparisons between the synthetic output and recordings of a selected model speaker. The oral presentation will emphasize differences between the English and Japanese synthesis systems. A demonstration will be played. [Work supported in part by an NIH grant.]