Syllabic Nuclei Research Articles

Attributes of lateral consonants

This paper attempts to define acoustic properties of laterals in English that are adjacent to vowels and in syllabic nuclei. When the tongue is in the lateral configuration, there is an increased bandwidth and decreased prominence of the second formant (due to increased losses) and some irregularities in the F3–F4 region (divided acoustic path). When a lateral consonant is produced adjacent to a vowel, there is a rapid change in the acoustic spectrum as these attributes appear or disappear. A series of acoustic analyses has investigated the nature of this discontinuity, particularly the change in amplitude and frequency of the first two spectral prominences. An abrupt change in F2 amplitude of about 10 dB at consonant release (less at closure) is a salient attribute. The attributes that distinguish a syllabic lateral and a nonlow back vowel /o/ have been examined through acoustic analyses and through perceptual experiments in which F2 bandwidth and other properties are manipulated in synthetic utterances. F2 bandwidth again appears to play a significant role in signalling this distinction. The common acoustic attributes associated with the lateral configuration appear to be used in different ways by listeners depending on the context. [Research supported in part by NIDCD.]

Neural network approach to assist markings of syllabic nuclei for multi-language database

The manual marking of syllabic nuclei is not difficult for a trained person. For a large amount of speech data, it is time consuming, and human error can introduce inconsistencies. The problem becomes worse when the person is not familiar with the language. To provide an automatic procedure for reducing the marking time and eliminate human errors, an adhoc algorithm was developed to find syllabic nuclei using a set of acoustic features. The initial error rate is 15%. The overall time spent marking automatically and correcting errors manually, however, was less than hand-marking by itself. An improved machine learning approach using a backward-error-propagation multi-layered perceptron, with the same acoustic features as inputs, was applied to the training speech. The hand-corrected markings was used as the teacher. Testing on the training data showed only 3%–5% misalignment. The time required to verify markings has further been reduced. When the training extended to a database of 43 different languages, the marking errors dropped to 2%, less than the errors in the initial hand-markings. Finally, when the network is used to mark 7 h of a multi-language database, there is no significant increase in marking errors.

Contingencies in stress patterns and syllabic nuclei: Lessons for searching lexical databases

In automatic speech recognition programs, attempts have been made to use partial information about the phonetic structure of signals to search large lexical databases. Little attention has been given to the interaction of stress patterns and syllabic nuclei in restricting the searches. A survey of the MRC Psycholinguistic Database of British English (RP) with 38‐K tokens revealed that six stress patterns (out of 17 attested) account for 96% of the tokens in the subset of two‐ and three‐syllable words. Further, three syllabic nuclei (out of 23 attested) account for 50% of the tokens, independently of stress pattern. The interactions of stress pattern and nuclei show not only increased discriminatory power (in information theoretic terms) over either attribute alone, but also reveal significant gaps between expected and observed cases. The gaps, both accidental or principaled, further reduce the relevant search spaces and increase the usefulness of this type of partial information over other candidates. [Work supported by Alvey Grant MMI 069.]

Recognitions of demisyllables based on dynamic programming methods

This paper describes the application of a one-stage Dynamic Programming (DP) algorithm to the acoustic-phonetic decoding stage in a speech recognition system. Essentially two methods are compared: (1) the recognition of demisyllable. and (2) the recognition of consonant cluster and syllabic nuclei (vowels or diphthongs). Demisyllables, consonant clusters and vowels or diphthongs have proved their usefulness in the recognition of continuous speech. Furthermore, the method of Dynamic Programming is a well established principle for the recognition of connected words. In this paper a stage of acoutic-phonetic decoding is presented which applies Dynamic Programming to these phonetic units. The algorithm used is essentially the same as the one commonly known for connected word recognition, which does not need explicit segmentation. However, for the recognition of demisyllables, consonant clusters and vowels, some specific modifications were necessary, including the introduction of a special internal syntax between the units. The derived algorithm was tested using fluently spoken sentences from 75 word lexicon in which the most frequent consonant clusters and vowels of the German language were represented. Different tests were made in order to get a comparison of the recognition accuracy using a context-dependent application (concerning neighbouring units) and a context-independent application. These results and a comparison to a similar acoustic-phonetic stage using an explicit segmentation are presented.

The Old Spanish Graphs ‘i’, ‘j’, ‘g’ and ‘y’ and the Develópment of Latin Gei- and J-

The Old Spanish Graphs ‘i’, ‘j’, ‘g’ and ‘y’ and the Develópment of Latin G^ei- and J-

Acoustic‐phonetic analysis of Spanish /I/

Acoustic measurements to examine the contribution of overlapping segments—anticipatory and carryover coarticulations—to the identification of American Spanish /I/ were performed. The consonant was combined with five vowels in naturally produced CV, VC, VCV, and CCV syllables. Acoustic profiles through analysis of formant trajectories, fundamental frequency, intensity and duration of vowel, consonant, and their overlapping segments were obtained. In searching for the corresponding critical spectral features, digitalized waveforms were modified by deleting various temporal segments in the syllable. The excised sounds were used as stimuli in identifications tests. Results indicate that the duration of the syllabic nuclei varies across front, medial, and back vowels and also in pre‐ and post‐vocalic position. Segment /I/ presented in isolation is poorly identified. Best identifications are obtained when the stimuli are integrated by the end of the overlapping portion and the beginning of the vowel. A trading off relationship effect is observed between these two segments.

The Phonology of Syllabic Nuclei in Slovak

The Phonology of Syllabic Nuclei in Slovak

The representation of phonological information during speech production planning:evidence from vowel errors in spontaneous speech

ABSTRACTA corpus of more than 500 speech errors that involve a vowel or syllabic nucleus is examined for evidence that bears on the nature of the processing representation that is in force when such errors occur. Evidence is obtained from the patterns of similarity between target segments and the intrusion segments that replace them in errors, on the assumption that target– intrusion similarity arises from characteristics of the processing representation. Findings include (1) a distinctive feature similarity between vowel targets and intrusions, (2) evidence that complex syllabic nuclei can function as error units and (3) evidence that vowel errors are constrained by lexical stress. Finally, the error patterns in both vowels and consonants, and the processing representations they suggest, are evaluated in the light of recent theoretical proposals about the phonological component of the grammar.

Comparison of acoustic features of word accent in English and Japanese.

In order to determine the universal and language-specific characteristics for wordaccent, its acoustic manifestations were analyzed and compared in the disyllabic words of English (“permit, ” “record, ” “object” etc.) and in the two-mora words of Japanese (“ame”). Analyses on the fundamental frequency contours (F0 contours) of these words were made using the functional model of F0 contour generation proposed by one of the authors. While a marked similarity was observed between F0 contour characteristics of English and Japanese in cases of both first-syllable accented and second-syllable accented, individual differences were much greater in the accent command for English words with an accented first syllable. Segmental and syllabic durations were measured on the speech waveform, and it was found that the accentual changes in duration occur mainly in the second syllable in Japanese, while in English they tend to be complementary in the first and the second syllables. The intensity and formant frequencies of syllabic nuclei were also analyzed. The results of these analyses indicated that duration, intensity and formant frequencies are less stable than F0 contour as correlates for the word accent both in English and in Japanese.

Fundamental frequency and its relationship to Spanish stress

In the present work, we attempt to describe F0 movements and the amount of F0 change in the stressed syllables of Spanish sentences as a function of phonetic and linguistic effects. The material consisted of a list of 92 sentences with different syntactic structures which were recorded twice by a male speaker. Wideband spectrograms and F0 contours of all the sentences were obtained. F0 movements on the syllabic nucleus of each accented syllable were chosen as the parameters for the analysis of F0 contours. Results showed that the accented syllabic nucleus presents the following F0 movements: rise, fall, level, and rise-fall. These movements are correlated with effects such as the nature of the preceding consonant (level with voiceless, rise with voiced), position of the accented word (fall in the last accented word, rise in yes-no questions), and boundary (rise-fall or fall). Emphatic stress is correlated with increased F0 and larger F0 excursions. The amount of F0 change seems to follow a declination line. This general pattern is altered by different effects: lexical (adverbs, pronouns), syntactic (phrasal verbs, aposition, prepositional phrases, clauses, syntactic phrases), and semantic (modality operators, emphasis, prominent words).

Testing Bolinger's hypothesis of pre‐accentual lengthening

Algorithms were developed for acoustic prosodic analysis (Lea and Clermont, Proc. 1984 ICASSP, paper 42.7), including voicing decisions, syllabic nuclei locations, phrase boundary detections, and stressed syllable locations. Other algorithms permit interactively entering listener decisions of syllabic durations, perceived stress levels, identifications of syllables (or other segments) that test specific hypotheses, and linguistic predictions. Using such algorithms and standard signal processing and statistics software, researchers can automatically and efficiently test prosodic hypotheses. Selected for initial application of such capabilities was Dwight Bolinger's unpublished hypothesis that any full syllable is longer (and of higher Fo) immediately before another full syllable than before a reduced syllable. Minimal‐pair contrasts in accent sequences were embedded in 50 sentences spoken at fast and slow rates by each of two talkers, and 45 all‐sonorant sentences spoken twice by one speaker. Syllabic durations were derived from listener markings and from automatic syllabification, then compared, and statistically analyzed, for verification of the hypothesis. Fundamental frequency values at centers of nuclei were also studied.

Portuguese nasalized vowels and nasal consonants: Their relation to following consonants

Pilot spectrographic results indicate that Brazilian Portuguese demonstrates a relationship similar to that found in English between syllabic nucleus duration and the voicing characteristic of a following obstruent: vowels lengthen before voiced obstruents and shorten before voiceless obstruents. Also, a nasal consonant in Portuguese acts in conjunction with a preceding vowel as a syllabic nucleus; for example, if in a syllabic nucleus a vowel does not show the duration expected before a given obstruent, the nasal duration compensates, and vice versa. New data from six speakers of Brazilian Portuguese derived from computer analysis will be presented which confirm the preliminary results. The acoustic characteristics of nasalized vowels and nasal consonants in Portuguese will also be described, as well as the extent to which a prenasal vowel is nasalized. The difference between phonemically nasal vowels and vowels nasalized because they are adjacent to a nasal consonant (e.g., [1χ] lã, “wool” vs [pʌ̃nᴜ] pano, “cloth”) will also be discussed.

Dynamic specification of coarticulated vowels.

An adequate theory of vowel perception must account for perceptual constancy over variations in the acoustic structure of coarticulated vowels contributed by speakers, speaking rate, and consonantal context. We modified recorded consonant-vowel-consonant syllables electronically to investigate the perceptual efficacy of three types of acoustic information for vowel identification: (1) static spectral "targets," (2) duration of syllabic nuclei, and (3) formant transitions into and out of the vowel nucleus. Vowels in /b/-vowel-/b/ syllables spoken by one adult male (experiment 1) and by two females and two males (experiment 2) served as the corpus, and seven modified syllable conditions were generated in which different parts of the digitized waveforms of the syllables were deleted and the temporal relationships of the remaining parts were manipulated. Results of identification tests by untrained listeners indicated that dynamic spectral information, contained in initial and final transitions taken together, was sufficient for accurate identification of vowels even when vowel nuclei were attenuated to silence. Furthermore, the dynamic spectral information appeared to be efficacious even when durational parameters specifying intrinsic vowel length were eliminated.

A syllable parsing algorithm for telephone‐quality speech

This paper describes an algorithm for the automatic parsing of continuous speech into a string of syllables, intended for use as the “front end” of a demisyllable‐based continuous speech recognizer. The algorithm takes an input an autocorrelation coefficient (ACC) parameterization of speech, an analysis independently required for our recognition process; it makes use of only the zeroth‐order ACC (energy) and normalized first‐order ACC. Local maxima in a severely low‐pass‐filtered representation of the energy signal determine potential syllabic nuclei. Some of these first‐pass syllables are discarded on the basis of further tests. Compared to other syllable parsers described in the literature, this algorithm (1) was designed and tested for use with telephone speech, i.e., utterances spoken into a carbon microphone over a dialed‐up phone line and bandpass filtered to 200–3200 Hz; (2) locates potential syllabic nuclei by reference to the simple energy signal, rather than to a spectrally weighted “loudness” t...

A cineradiograhic study of static and dynamic aspects of American English /r/.

High speed cineradiograms were used to examine the articulatory characteristics associated with selected allophones of American English /r/. Only one of the three speakers produced a 'retroflexed' /r/ with a raised tongue apex. The other speakers produced a 'bunched' /r/ characterized by a raising of the tongue dorsum. Two basic types of /r/ were observed for all three speakers: (1) prevocalic /r/, which initiated a syllable, and (2) post-vocalic /r/, which terminated a syllable or formed a syllabic nucleus. The prevocalic allophone was characterized by greater lip rounding, a more advanced tongue position, and less tongue dorsum grooving. The two allophones also were associated with different variability and movement attributes. The differences in /r/ allophones were found to be similar to previously reported differences in American English /l/ allophones.

A demisyllable inventory for speech synthesis

English syllables can be treated as a syllable core, decomposable into initial and final demisyllables, plus optional word‐final phonetic affixes (e.g., /s,z,t,d,θ/) [Lovins and Fujimura, J. Acoust. Soc. Am. 60, S75 (A) (1976)]. An inventory of 835 demisyllables, five phonetic affixes, and 100 reduced‐vowel units (CV or VC syllables) has been prepared in the form of LPC parameters. This inventory allows us to generate any of the ten thousand possible English syllables. An initial demisyllable is composed of an initial consonant or cluster, plus a fixed‐length segment which includes the CV transition; a final demisyllable contains the major portion of the syllable nucleus, plus postnuclear consonant(s). We have found it generally appropriate to produce “tense” or “long” syllabic nuclei by concatenations such as /Cɪ−/ + /−iyC/ For stressed syllables, this allows a reduction of the set of initial demisyllable vowels to /ɪ,ɛ,æ,a,ʌ,ɔ,ᴜ/. Final demisyllables distinguish all fifteen nuclei. Only two spectral smoothing rules are needed to derive syllable cores from concatenated initial and final demisyllables. Results of naive listeners' perceptual evaluation of a sample set of concatenated monosyllables will be discussed. We will also present a tape recording of sentences produced from the inventory.

Three types of diphthongization

Diphthongization of vocalic syllable nuclei is not a process of one kind. In Dutch, for instance, diphthongization can be of three different types, depending on whether the syllabic nucleus is a long vowel, a genuine diphthong, or a vowel + semivowel sequence. The perceptual and acoustic aspects of these differences are well documented in the literature. This paper complements the description with some articulatory data, based on electromyographic recordings of extrinsic tongue muscle activity in relation to the three types of diphthongization. [Work supported by NINCHDS grant NS 13870 and NFWO, Belgium.]

Vowel utterances of young infants

Syllable-like utterances produced by 14 different infants aged 1–40 weeks were presented to two trained listeners. The listeners first transcribed a portion of the data phonetically. They then compared transcriptions, attempted to resolve disagreements and decided upon conventions to be used. One hundred and twenty-two additional segments were then transcribed by the listeners independently. On 71% of these utterances the listeners disagreed in their transcriptions of syllabic nuclei. The remaining 29%, upon which they agreed, were all transcribed as mid or low front. Their judgments, where they disagreed, were always vowels adjacent to one another within the vowel triangle and were also predominantly heard as mid or low front vowels. The vowels at the extremes of the triangle, /a/, /u/, and /i/ were not judged to be present at all. Ten vowels upon which the judges agreed and ten upon which they disagreed were analyzed by conventional spectrographic and by computer-assisted spectral techniques. The results of these analyses will be discussed. [Work supported by NINCDS.]

A prosodically guided speech understanding strategy

Our strategy for computer understanding of speech uses prosodic features to break up continuous speech into sentences and phrases and locate stressed syllables in those phrases. The most reliable phonetic data are obtained by performing a distinguishing features analysis within the stressed syllables and by locating sibilants and other robust information in unstressed syllables. The numbers and locations of syntactic boundaries and stressed syllables are used to select likely syntactic and semantic structures, within which words are hypothesized to correspond to the partial distinguishing features matrices obtained from the segmental analyses. Portions of this strategy have been implemented and tested with hundreds of seconds of speech, involving fifteen talkers. A program for detecting syntactic boundaries from fall-rise patterns in fundamental frequency contours correctly detected ever 90 percent of all predicted boundaries. An algorithm for locating stressed syllables (from fundamental frequency contours and high-energy syllabic nuclei) correctly located the nuclei of over 85 percent of all those syllables perceived as stressed by a panel of listeners. A study of segmental analysis results obtained by several other research groups showed that phonetic recognition clearly is most successful in the stressed syllables. Procedures for classification of stressed vowels, location and classification of sibilants, and location of stops, nasals, and [r]-like sounds have been implemented. Prosodic aids to parsing and semantic analysis are being investigated.

An Algorithm for Locating Stressed Syllables in Continuous Speech

Local increases in fundamental frequency (F0) and large integrals of energy in the syllabic nucleus are known to be among the best acoustical correlates of stress. Major syntactic constituents have been shown to have archetype rapid-rise-then-gradual-fall F0 contours, with the rise into the maximum F0 often associated with the first stressed syllable in the constituent. An automatic procedure for detecting constituent boundaries and maximum F0 positions in constituents [W. A. Lea, “An Approach to Syntactic Recognition without Phonemics,” IEEE Trans. Audio'Electroacoust. AU-21, No. 3 (1973)] and sonorant energy and F0 functions, provided input data for an algorithm for locating stressed syllables. The first stressed syllable of a constituent was associated with a high-energy-integral portion near the rising F0 into maximum F0 position. Other stressed syllables were associated with high-energy-integral portions near local increases in F0 above a steadily falling “archetype line” from the maximum F0 position to the end of the constituent. For over 400 sec of speech, including written texts, and questions, commands, and declarations for man-machine interaction (involving 16 talkers), over 85% of all syllables perceived as stressed by a panel of listeners were correctly located.