Neil A. Macmillan

Resolution for speech sounds: Basic sensitivity and context memory on vowel and consonant continua

Discrimination and identification experiments were performed for a vowel continuum (/i/– /i/–/q/) and two consonant continua (/ba/–/pa/ and /ba/–/da/–/ga/). The results were interpreted in terms of a generalization of a theory of intensity resolution [N. I. Durlach and L. D. Braida, J. Acoust. Soc. Am. 46, 372–383 (1969)] that makes precise the distinction between basic sensitivity (sensory‐based resolution) and context coding (labeling processes). On the vowel continuum, basic sensitivity increased gradually across the range, whereas, for both consonant continua, sensitivity peaked between phonetic categories. All speech continua were found to have small ranges (measured in jnd’s); context memory was good, and better for consonants than for vowels. The stimuli that could be labeled most reliably were near the category boundaries on the vowel continuum, but near good phonetic exemplars for consonants. Introduction of a standard in identification primarily altered response bias, not sensitivity.

Probe‐signal investigation of uncertain‐frequency detection

Observers in a 2IFC detection task were instructed to listen for ’’primary’’ tones with one of two frequencies, in the presence of a broad‐band background noise. On a small proportion of trials, probe tones of other frequencies were presented instead, following the procedure of Greenberg and Larkin [J. Acoust. Soc. Am. 44, 1513–1523 (1968)]. The function relating percent correct to signal frequency may be interpreted as an average internal filter characteristic. All six observers responded correctly on 70%–90% of trials on which primary signals were presented; performance on probe trials was poorer, falling to chance for probe frequencies sufficiently above the higher primary, sufficiently below the lower primary, and for some probes between the two primaries. The widths of the listening bands around the primaries were found to be similar to the widths estimated from one‐primary control conditions. Examination of the sequential data suggests that, while the pattern of sensitivity changes from trial to tri...

Integrality in the perception of tongue root position and voice quality in vowels

In English and a large number of African and Southeast Asian languages, voice quality along a tense-lax dimension covaries with advancement of the tongue root in vowels: a laxer voice quality co-occurs with a more advanced tongue root. As laxing the voice increases energy in the first harmonic relative to higher ones and advancing the tongue root lowers F1, the acoustic consequences of these two articulations may integrate perceptually into a higher-level perceptual property, here called spectral flatness. Two Garner-paradigm experiments evaluated this interaction across nearly the entire range of tense-lax voice qualities and a narrow range of F1 values. The acoustic consequences of laxness and advanced tongue root integrated into spectral flatness for tenser and laxer but not for intermediate voice qualities. Detection-theoretic models developed in earlier work proved highly successful in representing the perceptual interaction between these dimensions.

Central and Peripheral Processes in the Perception of Speech and Nonspeech Sounds

In this chapter, we apply a psychophysical theory to several speech and nonspeech continua. Our approach meets many of Repp’s (this volume) criteria for psychophysics: it is auditory-based, methodological, and concerned with speech sounds. We deviate from Repp’s characterization, however, in that our theory is not psychoacoustic, that is, is not limited to relations between stimulus characteristics and sensations.

Learning non-native vowel categories

Two hypotheses have recently been put forward to account for listeners ability to distinguish and learn contrasts between speech sounds in foreign languages. Bests (1994) perceptual assimilation hypothesis predicts that the ease with which a listener can tell one non-native phoneme from another varies directly with the extent to which these sounds assimilate to different native phonemes. Pisoni et al. (1994) have argued that training listeners to identify non-native phonemes teaches them sets of exemplars rather than leading to the abstraction of more general prototypes. We report the results of four experiments examining how American English listeners learn to perceive the contrasts among the front rounded vowels of German. The results suggest that listeners responses are a function of the phonetic dissimilarity of the vowels themselves rather than their assimilability to American English vowels, a result incompatible with the strong phonological interpretation of Bests hypothesis, but compatible with the weaker category recognition interpretation. These results also show strong speaker effects, and are thus compatible with Pisoni et al.s exemplars-not-prototypes interpretation of non-native category learning.

Minimum audible angle for clicks with simulated echoes: Effects of azimuth and standard

Minimum audible angle (MAA) was measured in free field with a center position at 0°, and 25° or 50° on the right. Stimuli were 6‐ms‐wide band noise bursts presented as single bursts or noise pairs under precedence‐effect conditions. In the single‐noise condition, one burst was presented from the right or left. In precedence lead discrimination, a leading sound was presented from the left or right speaker and a lagging sound from a speaker at center. In precedence lag discrimination, the lagging sound was on the right or left and the leading sound was at center. The test stimulus was either presented in isolation or was preceded by a single standard noise at center. MAAs were lowest for the single condition, slightly higher for lead discrimination, and much higher for lag discrimination. Presence of a standard yielded a benefit in lag discrimination, especially at 0°. Thresholds at 50° were significantly higher than at 0° and 25°, especially in lag discrimination. The data can be described by a model in wh...

Acquisition of non‐native vowel categories.

Is the acquisition of foreign phoneme identification determined by native abstract categories [Best, Development of Speech Perception, pp. 167–224 (1994)] or by particular tokens [Pisoni et al., Development of Speech Perception, pp. 121–166 (1994)]? If Best is correct, members of all instances of the same phonological contrast should be equally easy to identify. If Pisoni et al. are correct, then listeners should not generalize to new speakers or contexts. The acquisition by American English listeners of the potentially four front rounded German vowels /y:,y,oe:,oe/ was examined. Speaker and consonantal context were manipulated. Listeners improved with training. Our listeners, in contrast to those of Pisoni et al. (1994), generalized to new speakers and contexts. Contrary to Best’s 1994 prediction that identification of the same contrast would be equally easy, it was found in this study that listeners who heard one set of speakers identified the mid tense:lax /oe:,oe/ pair more accurately than the high /y...

Laxness of voice quality integrates with F1 (usually, but not always, negatively)

Both lax voice and advanced tongue root typically concentrate energy at low frequencies: Laxing the voice increases the relative prominence of the first harmonic, and advancing the tongue root lowers F1. A previous study [Thorburn et al., 2871 (A) (1994)] using the Garner paradigm found that these acoustically similar effects integrated perceptually for vowels in CVC context. In this study, the range of voice quality was extended to include tensor values. Across the entire set of voice qualities sampled in the two experiments, laxness integrated negatively with F1 at the lax and tense ends of the continuum but positively at intermediate values. This pattern of mean integrality is distinct from the additional finding of variance integrality, that is, greater uncertainty in judging voice quality when F1 varied than when it did not (and similarly for judgments of F1). According to a model of Durlach et al. [Percept. Psychophys. 46, 293–296 (1989)], the variance‐integrality effects can be attributed to a sensory rather than a context‐coding source. [Work supported by NSF and NIH.]

Internal filter characteristics for uncertain‐frequency detection

Observers in a 2IFC detection task were instructed to listen for “primary” tones with one of two frequencies, in the presence of a broad‐band background noise. On a small proportion of trials, probe tones of other frequencies were presented instead, following the procedure of G. Z. Greenberg and W. D. Larkin [J. Acoust. Soc. Am. 44, 1513–1523 (1968)]. The function relating percent correct to signal frequency may be interpreted as an average internal filter characteristic. All six observers responded correctly on 70%–90% of trials on which primary signals were presented; performance on probe trials was poorer, falling to chance for probe frequencies sufficiently above the higher primary, sufficiently below the lower primary, and for some probes between the two primaries. The widths of the listening bands around the primaries were found to be similar to the widths estimated from one‐primary control conditions. Examination of the sequential data suggests that, while the pattern of sensitivity changes from tr...

Intensity discrimination of continuous noise in five paradigms

Observers were presented monaurally with continuous wideband noise and were required to judge in which of two temporal intervals the noise was more intense. Threshold intensity differences at each of six signal durations (20–640 msec) were estimated for five conditions: (I) background alone in one interval, background‐plus‐increment in the other; (D) background versus background‐plus‐decrement; (II) background‐plus‐increment versus background‐plus‐larger‐increment; (DD) background‐plus‐decrement versus background‐plus‐larger‐decrement; and (ID) background‐plus‐increment versus background‐plus‐decrement. An energy‐detection model would predict, for all these conditions, that a log unit change in duration should lead to a 5‐dB change in threshold. Instead, the threshold‐versus‐duration curves ranged in average slope from 6.1 to 1.8 dB/log unit, being steepest for condition ID, and shallowest for conditions using decrements. In addition, the “pedestal” conditions, II and DD, yielded shallower curves than the...