Band 43, Heft 1-3

Research into microvariations of fundamental frequency (F₀) of speech (intrinsic and co-intrinsic effects) carried out on a large number of languages has brought to light universal tendencies which raise a number of problems, both from the point of view of perception and from that of production, for the interpretation and modelling of intonation. After a survey of these problems, experimental data for French are presented, concerning the intrinsic fundamental frequency (IF₀) of oral and nasal vowels and the co-intrinsic effect (CF₀) of consonants on the onset and pitch point of the F₀ of the following stressed vowels. It is shown that despite considerable intersubject variation, IF₀ differences between the classes of close, open and nasal vowels are significant in both prosodic contexts analysed, but that variations within these classes were not significant (except for the nasal vowels). The principal source of variation for the co-intrinsic effect (as has been observed in many languages) was the opposition voiced/voiceless for the preceding consonant. This effect was greater at the onset of the vowel (4 STs) than at the pitch point where it was comparable to the IF₀ effect (1 ST). A relative independence of the IF₀ and CF₀ effects was noted which tend either to combine or to cancel each other out depending on the phonetic context. A preliminary attempt to incorporate these effects into a synthetic model of intonation, by factoring out observed F₀ contours into segmental and suprasegmental components, suggests that unlike IF₀ and CF₀ effects of consonants, which intervene at a fairly late level of phonetic implementation, IF₀ effects of vowels are quite probably programmed at a higher level.

The paper deals with the role of two types of segmentally conditioned fundamental frequency (F₀) perturbation in speech perception, i. e. the differences in F₀ between high and low vowels and the effect of prevocalic obstruent voicing on F₀. The literature relevant to this issue is reviewed, and the results of a supplementary experiment on F₀ as a cue to vowel identity are reported. On this basis three major points are made: (1) Although the pitch of vowels is influenced psychoacoustically by their spectral shape, this effect is too small to be of any significant importance in speech perception. (2) The two types of F₀ perturbation are not treated differently in the speech perception process, and (3) irrespective of type, the segmentally conditioned F₀ variation can function as a cue both to segment identity and to the identity of prosodic categories, depending on the actual demands upon the speech perception system.

This article deals with the acoustical and perceptual importance of F₀ perturbations in word-initial breathy CV syllables of Hindi. The first part is devoted to the acoustical analysis of F₀ onset, F₀ trajectory, average F₀ of the breathy and steady portion, and duration of the breathy portion. It is concluded from the results that (i) F₀ perturbations are a function of the place of articulation of the stop and/or tongue height and/or tongue position of the vowel; (ii) F₀ onset reflects the larynx height and/or the activity of extrinsic larynx muscles, and (iii) the F₀ trajectory represents the coarticulatory processes within the syllable. The importance of F₀ onset, F₀ trajectory, average F₀, and duration of the breathy portion for the perception of breathy stops is examined in the second part of this study. It is shown, for 8 CV conditions /É–^ha É–^he É–^hi É–^ho É–^hu b^hi d^hi g^hi/ that F₀ onset, F₀ trajectory, and the duration of the breathy portion play an important role in the perception of breathy CV syllables in Hindi.

Two experiments, each with three sub-tests, examine the influences on lenis/ fortis perception of both preplosive F₀ (flat vs. rising) in monotone utterance intonation and of natural vs. synthetic speech quality. The German word ‘leiten’ [la^etn], ‘to lead’, pronounced by a male and a female speaker [Experiments I and II, respectively), formed the basis for the test stimulus construction. The stop closure (= silence) was varied by the same 5 steps to yield a continuum from clear /t/ to clear /d/ in all tests. The data support the conclusion that a rising as opposed to a non-rising F₀ at the end of a preplosive vowel in a monotone utterance intonation increases /t/ perception. They also show that LPC synthesis quality is a strong additional cue for /t/, being stronger where it characterizes the whole utterance rather than only a small part of it. These influences on lenis/fortis perception are found to be consistent for the male and the female voice.

Two different models of the influence of the stop voicing on F₀ are identified. The more widely accepted of the two, called the rise-fall dichotomy, claims that F₀ falls after voiceless stops but rises after voiced stops, and that the direction of post-release F₀ is contextually invariant. The alternative model, referred to as the no-rise view, arose from recent production studies. It claims that the onset frequency of post-release F₀ is raised after all stops, though only relatively little if they are phonologically voiced. More importantly, it sees F₀ contours as a combination of segmental perturbations added onto a smooth underlying intonation contour. Consequently the direction of post-release F₀ depends not only on segmental phonetic features but also on the prosodic structure. In three perceptual experiments utilising computer-synthesised intervocalic bilabial stops, opposing predictions of the two models are tested by embedding falling, level, and rising perturbations in different intonational environments. In each experiment the predictions of the no-rise view are supported and those of the rise-fall dichotomy are falsified. The conclusion drawn is that investigations of segmental phonetics can give rise to misleading results unless the accompanying prosodic structure is also taken into account.

In the experiment reported here, the intrinsic fundamental frequency (IF₀) difference between high and low vowels was investigated in controlled prosodic contexts. Findings do not support the hypothesis proposed elsewhere that IF₀ differences increase as a function of the speaker’s F₀. An explanation which includes the influence of both vocal fold tension and subglottal pressure accounts for the data better than current theories of IF₀ differences. IF₀ differences may be explained as a phonetic-level result of the implementation of pitch accent and prominence.

When listeners process segmentally relevant properties of the speech signal they do so in a rate-dependent manner. This is seen as a shift in the perceptual category boundary; as rate declines and overall syllable duration increases, the category boundary moves toward a longer value of the acoustic property in question. Focusing on the /b/-/p/ distinction specified by voice onset time (VOT), we investigated whether the acoustic modifications that occur with an alteration in speaking rate accord with this pattern of perceptual boundary shift. Two main findings emerged. First, as speaking rate became slower and overall syllable duration became longer, the VOT value of the consonant, especially that of the voiceless /p/, also became longer. Second, and most important, the VOT value that optimally separated the /b/ and /p/ VOT distributions also changed with rate, increasing with increasing syllable duration. However, the magnitude of the boundary shift obtained for these production data was greater than that typically found in perceptual experiments. This suggests the existence of constraints on the extent to which the perceptual system can accommodate for alterations due to rate of speech.

Numerous studies, both in speech production and perception, have found that the acoustic representation of speech segments is heavily context-dependent. In particular the durations of speech sounds do not have fixed values but vary depending on their immediate context, speech tempo etc. In spite of this variability, the perceiver usually perceives correctly what the speaker intended to say. Two theoretical accounts have been given of this fact. On the one hand it has been postulated that the listener perceives correctly by taking into account the context of the speech segment he hears (the theory of normalization). Another view is that speech sound variability is primarily a surface phenomenon, and that invariants can be found in the speech wave on a closer analysis (the theory of constancy). The present study investigates the merits of these theories in the context of Icelandic speech perception. The experimental results show that normalization plays only a minor role in the perception of quantity in Icelandic.

The durational properties of consonants have been studied for Swedish and English. The use of quantity in Swedish demands an expansion of the rule structure proposed by Klatt. The Swedish study has been done in the context of a speech data base of read sentences. There are considerable problems in administering such data bases, both to transcribe and segment the speech and to easily access stored material. We report how speech is transcribed and segmented semiautomatically. The transcription is generated by the text-to-phonetic component of our synthesis system. The same rule structure is used for accessing the data. Durational data is collected and compared to models directly during the data base search.