Allophonic Variation Research Articles

Speech synthesis from text with an articulatory model

In a film, a talking articulatory model describes itself, and its way of generating spoken English from text. Topics that may be considered include physical properties of the model, articulatory dynamics, sequencing of articulatory commands, relationships between articulatory commands and linguistic properties of sounds, and intermediate‐ and higher‐level prosodic effects, including segment duration, allophonic variation, boundary marking, stress, and intonation.

Syllabic and phonemic representations for short-term memory of speech stimuli.

An experiment is reported which uses a same-different matching paradigm in which subjects are required to indicate whether the consonants of a pair of consonant-diphthong syllables are the same or different. The question addressed is the operation of two hypothesized processes in the perception of speech sounds. The auditory level is shown to hold stimulus information for a brief period of time and be sensitive to allophonic variations within a stimulus. Moreover, matching at this level takes place by identity of the syllables rather than of the separate phoneme segments. The phonemic level is impaired when the diphthong segments of the pair leads to a contradictory match to that of the consonants of the pair, even though only the consonants are relevant to the matching decision.

Effect of allophonic variation upon phonetic feature recognition in dichotic backward masking

Synthetic stimuli (generated at Haskins Laboratories) were used to elicit monaural identification and discrimination data for two seven-step place continua /ba-da; pa-ta/, and two seven-step voicing continua /ba-pa; da-ta/. Five subjects who demonstrated categorical perception participated in three dichotic experiments. Using the same stimuli, dichotic experimental tapes with stimulus onset asynchonies ranging from 0 to 200 msec were made. Subjects were instructed to identify the stimulus presented to the right ear, which was always the leading syllable and a canonical form (step one or seven of the four continua). The first experiment employed canonical forms as the masker. Experiment two used stimuli from steps two-six from the voicing continua whereas experiment three used corresponding stimuli from the place continua. The target intelligibility data from experiments two and three were compared with data collected with data collected from experiment one and the monaural experiment. The influence of acoustic versus phonetic competition in the dichotic backward masking paradigm is discussed. [Work supported in part by NIH and Sigma XI.]

Word juncture perception as a function of word-initial phonemes

A previous study showed that allophonic variations in word-initial phonemes were the primary cues for word juncture perception. This study examined juncture perception as a function of the word-initial phonemes occurring singly and in clusters. Pairs of two-word phrases contrasting in juncture location (e.g., “he praises” and “heap raises”) and representing a wide variety of word-initial vowels, consonants and consonant clusters were recorded in sentence contexts by four talkers. The phrases were excised from the sentences and played to listeners who indicated which phrase they heard. Juncture perception was good for word-initial voiceless stops, vowels, affricates, and nasals, and fair for fricatives and voiced stops. Juncture perception was poor for liquids and semivowels, but good when they occurred in a consonant cluster. Juncture perception for clusters beginning with /s/ was better than for /s/ alone. Juncture perception was poor when the same consonant occurred before and after the juncture (e.g., “heap praises”); this suggests that segmental duration is not an important cue for juncture perception.

Coarticulation resistance in English /l/

Allophonic variations in the quality and voicelessness of RP /l/ were investigated, largely through their acoustic correlates as revealed by spectrography. The conventional separation of two distributionally-conditioned extrinsic allophones “clear” and “dark” was in general confirmed, although under different right-boundary conditions clear [l] was more widely used, itself varying in degree of clarity, than has been suggested elsewhere. Additionally, and more significantly, in each context /1/ varied further under coarticulatory influence—in its quality attributable to an adjacent vowel, in quality (in terms of overall clarity/darkness) attributable to an adjacent second lateral, and in voicelessness attributable to an adjacent voiceless consonant. Both right-to-left and left-to-right coarticulation effects were observed extensively: surprisingly, both types seemed constrained by a CV -type articulatory syllable. One general principle emerged which better explained the pattern of coarticulatory variation in the data: coarticulation resistance, a property which it is proposed to associate with phonetic specifications for speech segments in the form of values whose magnitude varies, increasing in the case of RP [ł], [ł] and [l] in that order. The validity of this principle for these and perhaps for other published data is independent of the direction (R-L or L-R) of the coarticulatory effect, and of the feature (vowel-quality, lateral-quality or voicelessness) being coarticulated.

Vocabulary and syntactic complexity in speech understanding systems

Comparing the realtive performances of speech-understanding systems has always been difficult and subject to speculation. Different tasks naturally require different vocabularies with varying acoustical similarities. Moreover, constraints imposed by the syntax may make recognition easier, even for vocabularies with high ambiguity. We define “inherent size” as a measure of vocabulary complexity and investigate its relation to recognition rates and the (apparent) vocabulary size. Word recognition is modeled as a probabilistic function of a Markov process using phoneme confusion probabilities derived from an articulatory position model. Multiple pronunciations and allophonic variations are allowed. Analysis of maximal word confusions leads to lower bounds for expected word recognition rates. Inherent vocabulary size is derived from these bounds. To evaluate syntatic constraints for finite state languages, an inherent size is computed for each state based on the subset of words which lead from that state. Then, average path complexity is computed for the language. This language complexity, when compared with the total vocabulary complexity and average path length, yields the degree of syntactic constraint. Analysis of several tasks will be reported.

Allophonic variations of stop consonants in a speech synthesis-by-rule program

In natural speech the acoustic form of each consonant is affected by its context, especially by the vowel which follows it. A computer program for synthesizing speech by rule from phonetic data has been modified so that the rules for generating stop consonants depend upon context. Listening tests have shown that this expedient can increase the intelligibility of stop consonants in isolated CV syllables from 68 % to 92%.

Cinefluorographic study of selected allophones of English /1/.

Abstract The purpose of this cinefluorographic study of a number of allophonic variations of English /l/ was to describe the articulatory characteristics of these variations and to assess factors which may be responsible for the variations. Three subjects produced a speech sample containing English words and phrases while the high speed cinefluorographic films were made. The observed variability centered around two allophone groups, the prevocalic and the post-vocalic, having different articulatory characteristics. The pre-vocalic productions showed a more anterior lingual position and faster speed of movement than was observed for the post-vocalic group. Consequently, the pre-vocalic allophones showed no undershoot of the anterior lingual contact, while the post-vocalics did show undershoot of this position. The syllabic /l/ productions displayed lingual configurations like the post-vocalic group and movement characteristics like those shown when /l/ is articulated in a pre-vocalic cluster. Variation within the three major allophone groups can be explained within current concepts of coarticulation and articulatory undershoot.

The perception of stop-liquid clusters in phonological fusion

Phonological fusion occurs when items such as PAY and LAY are presented separately to each ear and listeners report hearing PLAY. Input items that begin with a stop consonant (e.g. /p/) and a liquid (/1/ or /r/) fuse especially well. The present studies examined the effect of various factors on the frequency of fusion responses. Allophonic variation in the liquids (“trilled” versus plain) had no effect on fusion frequency. Phonemic similarity also had no effect; that is, when the input items differed in all phonemes (PAY and LED) they still fused. However, the phonemic order and location of clusters within a syllable did have a large effect:initial stop-liquid clusters fused readily (PAY/LAY yielded PLAY) while final liquid-stop clusters rarely fused (PEEL/PEED rarely yielded PEELED). Various fusion phenomena remained the same when tested in both identification and discrimination paradigms. Finally, fusion scores were not normally distributed over subjects; that is, some subjects fused on most or all trials, while others fused less frequently.

Allophonic variation in American English

It is well-known that two extremes of allophonic variations of voiceless stops serve as a cue to location of word boundaries. For example, the contrast between strongly aspirated and unaspirated /t/ is the primary difference in the utterances “grey tie” and “great eye”. If indeed such strong allophonic contrasts can be found in the few minimal pairs of word boundary that exist in English, it is reasonable to suspect a systematic pattern of variation of all consonants. Here we report a study of the consonants /t/, /d/, /n/, /l/ and /s/ from more than 20 min of reading, primarily by one speaker. We consider the devoicing time of /t/ and amplitudes of the other consonants, as functions of pausal, word, syllable and similar boundaries. The results show that the voiced portion of consonant tends to be low in amplitude and more consonantal in initial positions, and high in amplitude and more vocalic in final positions. The fricative portion of a consonant is more intense in initial positions and weaker in final ones.

Automatic synthesis from ordinary english test

We summarize work between 1969 and 1972 in a continuing project With two objectives: to produce acceptable synthetic speech directly from English text; and to demonstrate with speech synthesis a detailed model of human articulatory movements. Work in the four-year period has yielded moderately accurate rules for predicting the occurrence of pauses and lesser breaks in the sentence; rules for vowel duration in many conditions, not just primary stressed syllables immediately before a pause; rules for contextual variations of consonants; and rules for durational and other allophonic variations on consonants at word boundaries. Presently we are studying natural speech to quantify and add detail to these rules, and we are working to extend the vocal tract model to closer agreement with human articulation and vocal cord control.

Computer-Assisted Phonetic Analysis Techniques for Large Recorded Corpuses of Natural Speech

A system of computer programs has been developed for creating, from surreptitiously recorded conversations, data files containing phonetic transcription, intonation and stress patterns, allophonic variations, grammatical functions, demographic information such as the age and occupation of each speaker, psychological factors such as degree of formality, emotional state and intellectual level of subject matter, and paralinguistic factors such as syllabic rate, loudness, etc. The system also includes programs for locating phonetic patterns, expressed in a language readily understood by phoneticians with no computer training, and for analyzing them statistically, either for the corpus as a whole or as functions of any of the nonlinguistic or paralinguistic variables mentioned above. These techniques have initially been applied to 50 half-hour recordings of conversations with members of the Paris “establishment,” revealing a number of phonetic laws hitherto unsuspected. The method is now ready and available for the study of the phonology of other languages, and other applications have appeared. One particularly promising possibility is the establishment of patterns of phonetic habits for a given number of individuals for voice verification purposes.

Fricatives—Their Physical Properties and Allophones

Voiceless fricatives in American English were studied from readings of a continuous text. The wide range of duration (30 msec to more than 200 msec) is explainable in terms of phonological conditions. For example, a fricative in the stressed-syllable initial position is about 20% longer than in the same phoneme sequence in unstressed position; single fricatives are longer than those in clusters with the same stress and positional conditions; a fricative preceded by a nasal is about 30% shorter than one under equivalent conditions with a nonnasal. Overall intensity of fricatives is not simply proportional to duration, but rather depends on stress and boundary conditions. Fricatives in prepausal position are longest but weakest. For a given phonological condition (e.g., word-initial stressed, and single), overall intensity reflects the sentence stress. There are consistent relationships between fricative intensity and spectral distribution. Recently we have argued that the stop consonants vary systematically to mark, among other things, word boundaries and the beginnings of stressed syllables. This study shows that allophonic variations of the fricatives carry the same information.

On Significance of Allophonic Variation in Voiceless Stops; Some Results of Vocal-Cord Instrumentation and Speech Synthesis

A study of allophonic variations of voiceless stops in American English was made by transillumination of the glottis and by modeling in speech synthesis. Variants of phonemes are largely ignored in linguistic studies, on grounds that they are nondistinctive—i.e., insignificant in distinguishing the words that a phoneme string represents. In speech synthesis, however, we find the selection of allophones more crucial. Durations of closure and aspiration time of consonants in otherwise identical phoneme sequences can indeed force clear choices between such phrases as “the spit/this pit,” “at all/a tall,” etc. In measurement of vocal cord area by transillumination, data on the consonants /p,t,k/ show the glottis remaining open wide during the plosive burst and most of the aspiration of strongly aspirated stops, then closing rapidly to begin voicing. In weakly aspirated stops, the cords are kept closed, or nearly so, as if there were simultaneously articulated glottal stops. In stressed-syllable-initial /st/, the cords open wide during /s/ but close during /t/, reaching closure in time to suppress the burst and aspiration. Since the cords are apparently capable of opening or shutting in 30 or 40 msec, variants from 0 to 100 msec in aspiration time must be expressly controlled rather than accidental. Thus we see aspiration-time allophones as systematic (Monsen, Molter, and Umeda, previous abstract), explicitly selected and important for distinguishing word sequences. Whether or not they are nondistinctive, these allophones are not linguistically insignificant.

Suffixal j in Germanic

It is well known that Germanic nouns and verbs with a j-formative have two types of syllable structure. The suffix is said to represent /ij/ following a heavy or unstressed syllable, and /j/ elsewhere. Sievers (1878:129-30) noticed the allomorphic distribution of suffixal j for monosyllables, but was unable to give any rules for polysyllabic stems. He based his formulation on Germanic evidence, primarily on Old English, showing that it was necessary to postulate syllabic and consonantal variants in order to avoid contradictions with existing syncope rules for unstressed vowels (132-6). Research following Sievers' classic article has concentrated almost exclusively on the comparative aspect by linking the Germanic phenomena with allophonic variations in the Rigveda (Edgerton 1943).2 Regardless of the situation in Indo-European, it is necessary for Germanic to distinguish between a syllabic and consonantal variant of suffixal j. Conclusive evidence is found in Old English, e.g. in the nom.acc. pi. of heavy ja-neuters as compared with their light counterparts:3 *kun-j-u 'races', *rzk-j-u 'kingdoms'. In the attested nom.acc. pi. cyn, ricu one observes loss of final u after a light stem syllable, but not after a heavy one. This contradicts the apocope rules for Old English: (1) in monosyllabic stems, high vowels drop after a light syllable; (2) in polysyllabic stems, final u drops in forms of the structure

Generative Theory of Segmental Duration in English

This paper presents a framework for predicting the durations of phonetic segments in English sentences. The rules have been developed to work within the transformational cycle described by Chomsky and Halle. It is assumed that the rules of word-level phonology produce a set of basic phonetic segments, each having an inherent duration. Stressed and unstressed vowels are assigned different durations and are subject to different rules. Consonants appearing in prestressed position within a word are given durations that are longer than the durations of the same consonants in other phonetic environments. Other rules shorten segmental durations in consonant clusters, in multisyllable words, and in certain syntactic environments. Silent pauses are inserted at some types of syntactic boundaries and a prepausal lengthening rule is applied. Other related phonological processes involving aspiration and allophonic variation are discussed. [This work was supported in part by the National Institutes of Health and the U. S. Air Force.]

An allophonic problem in Australian English.

The phoneme /ou/ normally manifests itself with a consistent difference as between Non-Broad and Broad Australian English but yields before /I/ an allophone that is common to both speech varieties. The spectrographic investigation described below indicates that the starting-point of this allophonic glide has been retracted and lowered to about the / / position. A perception test has revealed a marked disinclination on the part of speakers of Australian English to accept the glides in such words as so and soul as the same "sound" i.e. to respond in the normal fashion to allophonic variation. Nonetheless economy in procedure seems to preclude a resultant phonemic re-writing of Australian English.

The rephonemicization of Modern Greek

It is possible by means of analysis into simultaneous components to reduce the number of phonemes normally posited for Modern Greek to twenty. Various methods of treating the semiconsonant [j] are discussed and the voiced stops are shown to be allophonic variants of the voiceless series. The ‘affricates’ can more conveniently be regarded as clusters.