John W. Hawks

A formant bandwidth estimation procedure for vowel synthesis

The specification of vowel formant bandwidths for speech synthesis has been inconsistent in the past, perhaps due to the difficulty of measuring formant bandwidths in natural speech and the possible perceptual insignificance of formant bandwidths on the intelligibility of synthetic speech. Here, regression equations are presented for the estimation of formant bandwidths based on measurements from natural speech which is based only on formant center frequency and independent of other formant values. Current usage, as well as comparison with another well-known estimation algorithm suggests that the new procedure should be quite acceptable for some types of speech synthesis.

Difference limens for formant patterns of vowel sounds

Studies of thresholds for discrimination of formant frequency variation in synthetic vowel sounds have been predominantly limited to variations in a single formant. Here, differences limens (DLs) are presented for multiformant variations expressed in measures of delta F and as distances in the auditory-perceptual space (APS) proposed by J. D. Miller [J. Acoust. Soc. Am. 85, 2114-2134 (1989)]. DLs for four subjects were estimated along 102 synthetic vowel continua representing five patterns of formant variation [(1) single variation in F1; (2) single variation in F2; (3) parallel simultaneous variation in F1 and F2; (4) opposing simultaneous variation in F1 and F2; and parallel simultaneous variation in F1, F2, and F3] and 17 within- or between-category vowel sounds. Minimal uncertainty methodology was employed utilizing an adaptive up-down procedure with a cued, two-interval forced-choice (2IFC) task. The results of this experiment reflect smaller DLs for both single- and multiple-formant changes than have been found in the past and also suggest that discrimination of parallel multiformant variation is significantly better than opposing multiformant or single-formant variation.

Bi-syllabic, Modern Greek word lists for use in word recognition tests

The development of a word recognition test for Modern Greek, which is comprised of three fifty-word lists, is described herein. The development was guided by four principles: 1) use of the shortest words possible (two syllables for Greek) 2) use of highly frequent words 3) phonetic balance and 4) appropriate balance of first and second syllable stress. The lists were recorded by one male and one female native speakers. Thirty-seven native speakers of Greek listened to all words by both speakers. Across lists, the mean correct identification score was 97.9% for the female voice (95% confidence interval 96.97 to 98.84) and 96.5% (95% confidence interval 95.31 to 97.77) for the male voice. This small difference was statistically significant (p<.01) and concentrated on words with first syllable stress. In future work, these recordings can be used in adult tests of speech perception and can be modified for tests of central auditory processing. Sumario Se describe el desarrollo de una prueba de discriminación en griego moderno a partir de tres listas de cincuenta palabras. Este desarrollo se basó en tres principios: 1. Uso de las palabras más cortas posibles en Griego (bisilábicos); 2. Uso de palabras altamente frecuentes; 3. Balance fonético 4. Balance apropiado entre el acento inicial y final. Las palabras fueron grabadas por un hombre y una mujer nativos del idioma. 37 personas nativos del idioma griego escucharon todas las palabras con ambas voces. En todas las listas, el porcentaje de identificación correcto fue del 97% para la voz femenina (Intervalo de confianza de 95% de 96.97 a 98.84) y 96.5% (intervalo de confianza del 95%, de 95.31 a 97.7%) para la voz masculina. Esta pequeña diferencia fue estadísticamente significativa (p<.01) y se concentró en las palabras con acento en la primera sílaba. Estas grabaciones pueden ser utilizadas en futuros trabajos para pruebas de percepción del lenguaje en adultos y pueden ser modificadas para pruebas de procesamiento central auditivo.

Effect of Frequency Boundary Assignment on Speech Recognition with the Nucleus 24 ACE Speech Coding Strategy

The choice of frequency boundaries for the analysis channels of cochlear implants has been shown to impact the speech perception performance of adult recipients (Skinner et al, 1995; Fourakis et al, 2004). While technological limitations heretofore have limited the clinical feasibility of investigating novel frequency assignments, the SPEAR3 research processor affords the opportunity to investigate an unlimited number of possibilities. Here, four different assignments are evaluated using a variety of speech stimuli. All participants accommodated to assignment changes, and no one assignment was significantly preferred. The results suggest that better performance can be achieved using a strategy whereby (1) there are at least 7-8 electrodes allocated below 1000 Hz, (2) the majority of remaining electrodes are allocated between 1100-3000 Hz, and (3) the region above 3 kHz is represented by relatively few electrodes (i.e., 1-3). The results suggest that such frequency assignment flexibility should be made clinically available.

The perceptual vowel spaces of American English and Modern Greek: a comparison.

Perceptually based vowel spaces are estimated for American English and Modern Greek by means of identifications of synthetic vowel sounds by native speakers of each language. The vowel spaces for American English appear to be organized in a sufficiently contrastive system, while Modern Greek vowels appear to be maximally contrastive. The spaces for the Modern Greek point vowels ([i], [a], [u]) fall within the spaces of their American English counterparts, while the intermediate Modern Greek vowels ([e], [o]) overlap the American English [∊]/[e] and [[UNKNOWN]]/[o] spaces, respectively. These results were relatively unaffected by mapping resolution and level of phonetic training and support the results of similar mappings using production data.

Effects of formant bandwidth on the identification of synthetic vowels by cochlear implant recipients

Objective: The main objective was to investigate whether the broadening and narrowing of formant bandwidths had a significant effect on the identification of vowels often confused by Nucleus cochlear implant recipients using the Spectral Peak (SPEAK) speech coding strategy. Specifically, identification performance for synthetic vowels with the first two formants (F1 and F2) parametrically varied in bandwidth was explored. Design: Eight implanted subjects identified synthetic versions of the isolated vowel sounds [&Igr;, ε, &U028C;, &U025D;] with F1 and F2 bandwidth manipulations, as well as foil tokens of [i, u, &U0251;, æ, &U028C;]. Identification performance was examined in terms of percent correct as well as error patterns. Further analyses compared patterns of electrode activation. Results: In general, broader F1 bandwidths yielded poorer performance and narrower F1 bandwidths yielded better performance relative to identifications for the reference stimuli. However, similar manipulations of F2 bandwidths resulted in less predictable performance. Comparison of electrode activation patterns indicated a distinct sharpening or flattening in the F1 frequency region for subjects with the greatest performance extremes. Conclusions: Manipulation of F1 bandwidth can result in concomitant changes in electrode activation patterns and identification performance. This suggests that modifications in the SPEAK coding strategy for the F1 region may be a consideration. Similar manipulations of F2 bandwidth yielded less predictable results and require further investigation.

Perception of synthetic vowels by cochlear implant recipients

Adult recipients of cochlear implants (Nucleus and Clarion) provided identifications and confidence ratings for two sets of 465 vowel‐like tokens which were synthesized with two fundamental frequency (F0) contours representative of a typical male voice and female voice. Formant combinations used for the stimulus sets sampled a large range encompassing all nonretroflex vowels of American English. Subjects assigned each token to one of ten American English vowel qualities or to a category indicating it was not one of the ten, and rated their confidence of identification on a five‐point scale. Various perceptual maps were constructed from these responses based on number of agreements per token, most often used ID category per token, and average ratings. These maps will be compared with similar maps derived from normal hearing subjects. Comparison and discussion will address general differences in perceptual vowel spaces and rating patterns between the subject groups, as well as specific differences related t...

An investigation of the perceptual magnet effect in adults

Three experiments were undertaken to investigate the validity and strength of the perceptual magnet effect in adults. First, the original perceptual magnet experiment [P. K. Kuhl, Percept. Psychophys. 50, 93–107 (1991)] was directly replicated with adults, using Kuhl’s stimulus parameters. These subjects were then tested with the AAXX protocol used to test monkeys in Kuhl (1991). Generalization scores were computed for subjects under both conditions and compared. A second group of subjects rated a large number of /KY/ vowel tokens in the next experiment. Average ratings were computed and a prototype (P) and nonprototype (NP) were chosen based on them. Variants of each were synthesized as in Kuhl (1991). Subjects rated the goodness of each variant on a scale from 1 to 7. In addition they assigned a 0 to any token perceived as not belonging to the /i/ category. The subjects then participated in the perceptual magnet experiment. Data from trials containing stimuli assigned 0 were excluded from analysis. In t...

Perceptual overshoot with speech and nonspeech sounds

One of the basic quests in speech perception research has been to find the differences or similarities in the mechanisms involved in the perception of speech and nonspeech sounds. The current study will address the differences in perception of speech and nonspeech signals by comparing the perceptual overshoot in synthetic vowels and sinewave acoustic replicas of the synthetic vowels. Lindblom and Studdert‐Kennedy (1967) demonstrated that the perceptual boundary for steady state vowels and that for vowels in a CV context with F2 transition are different. They called this phenomenon a perceptual compensation or perceptual overshoot. In the current study the perceptual boundaries for synthetic steady state vowels, steady state sinewave acoustic replicas of vowels, a vowel in the CV context with F2 transition and sinewave acoustic replicas of vowels in the CV context are compared. The results will be discussed in the poster. For Speech Communication Best Student Paper Award.

Perception of vowel stimuli synthesized with different fundamental frequencies

This research examined the perception of 465 vowel stimuli synthesized with both a typically male and a female fundamental frequency (F0) by persons with normal hearing, testing the hypothesis that when people listen to male and female speakers they employ different criteria for each speaker in order to decide which vowel the speaker has produced. The subjects, five male and five female native speakers of Midwestern American English, listened to the 465 vowel tokens synthesized with both F0’s. They were asked to decide, for each token, if it was one of eleven possible vowels as in the words: heed, hid, hayed, head, had, hod, hawed, hoed, hood, who’d, hud, or if not, to respond NONE. Next, the token was played again and the subjects indicated their certainty on a scale of 5 (very certain) to 1 (very uncertain). The responses were used to create perceptual maps in which each synthesized point is represented by the vowel code assigned to it by the majority of the subjects. The resulting maps indicate that ma...