Phoneme Duration Research Articles

A model of segmental duration for speech synthesis in French

This paper presents a set of rules to predict phoneme durations for synthesis applications in French. The rules use a speaker-independent Intrinsic Duration for each phoneme and a lengthening/shortening coefficient reflecting the effects of context and speaking style. The model can thus yield different sets of phoneme durations as produced by different speakers. The validity of the model was tested on 2 speakers. For the test-corpora, the mean differences between predicted and measured durations were less than 18 ms.

Age‐related differences in phoneme durations among females

Reduced speech rate has been consistently reported for geriatric speakers when compared to the speech of younger adults. Only recently have the specific aspects of the speech signal associated with reduced rate been identified. The purpose of this study was to examine vowel and consonant durations using certain physiologic measures to determine their effect on age‐related differences in speech rate. Two groups of 25 women—one 20–30 years old and another 75 years and older—who had no history of speech disorders and exhibited normal hearing (for age level) served as subjects. They repeated a series of CV, VCV, and VC syllables using the consonants /p/, /b/, /t/, /d/, /s/, and /z/ combined with the vowel /ɑ/, which were spoken in the carrier phrase “Speak 3♯ again.” Intraoral air pressure and the voicing signal were recorded to provide vowel and consonant temporal information. Consonant duration did not differ significantly between the two age groups. Conversely, vowel durations were greater for the older females. These results are in agreement with previous studies and will be discussed in relation to a general model of aging.

Aging speech: Physiological and sociological aspects

It is generally accepted that listeners are able to accurately classify a speech sample according to the chronological age of the speaker (Ptacek and Sander, 1966; Shipp and Hollien, 1969; Ryan and Burk, 1974; Ryan and Capadano, 1978). Impressive listener accuracy has been reported in age ratings of both connected speech (Ptacek and Sander, 1966) and sustained vowel phonation (Shipp and Hollien, 1969; Linville and Fisher, 1980). Findings such as these suggest the acoustic characteristics of speech change with advancing age (Kent and Burkard, 1981) and that listeners use certain of these acoustic characteristics in their judgments of age. For example, sustained vowel phonation of the elderly has been described perceptually as hoarse, breathy and tremorous (Ptacek and Sander, 1966; Hartman and Danhauer, 1976; Ryan and Capadano, 1978); acoustic analyses of such sustained phonation have revealed increases in fundamental frequency, jitter, shimmer and spectral noise (Wilcox and Horii, 1980; Ramig and Ringel, 1983; Ramig, 1983a). Connected speech of the elderly has been reported to be slow, have long pauses and hesitancy and be less clear (Ptacek and Sander, 1966; Hartman and Danhauer, 1976); acoustic analyses of such speech reveal increased silent time, longer pauses and increased duration of phonemes (Mysak, 1959; Hoflien and Shipp, 1972; Hartman and Danhauer, 1976; Schow ef al., 1978).

Analysis of phoneme and pause durations in conversational and clear speech

When speech is spoken “clearly” the durations of segments and words increase relative to “conversational” speech, and there is a corresponding drop in average speaking rate. [Picheny, Durlach, and Braida, J. Acoust. Soc. Am. Suppl. 1 67, S38(1980)]. We report new measurements of pause durations (silence gaps exceeding 10 ms between words) and phoneme durations for content words in 50 nonsense sentences as a function of speaking mode. Separate statistical analyses were made for conversational durations, clear durations, and the percentage increase of durations. Results show that the percentage increase is strongly phoneme dependent: It varies from 15% for voiced plosives to 95% for semivowels. Analysis with respect to factors such as stress, serial position, and environment suggests that different production rules are used in the two speaking modes. [Work supported by NIH.]

A multispeaker analysis of duration in read French paragraphs

An understanding of natural language durations will lead to more intelligible synthetic speech, and to improved automatic recognition. Toward this goal, a I I I‐word paragraph was read by 29 native French speakers. A pitch‐extractor device plotted the speech amplitude waveform, at an average rate of 55 mm/s. Durations were measured to the nearest 5 ms, with a measurement error of about 10 ms (1/2 mm). Phoneme durations were found to be significantly shorter than those in stressed words from sentences pronounced in isolation [D. O'Shanghnessy, J. Phon. 9, 385–406 (1981)]. Previous trends (short schwa vowels and grammatical words, long unvoiced fricatives, nasal vowels, and prepausal syllables) were confirmed. Vowels lengthened preceding voiced fricatives (but not prior to /r/, and lengthened more at sentence‐internal pauses than at the end of a sentence. Standard deviations of phoneme durations (at a fixed position in the paragraph) across speakers averaged about 20% for consonants and 25% for vowels. A generative model of French durations will be presented. [Supported by NSERC and the France‐Quebec exchange.]

Prediction of phoneme duration by a distinctive feature matrix

For the analysis and synthesis of intonation, we tried to find a simple method of estimating intrinsic segmental duration. Multiple linear regression analysis was carried out using phoneme durations as the dependent variable with several types of independent variables: formants, a numerically coded vowel chart, and a binary coded distinctive feature matrix. We found that the distinctive feature matrix was suitable not only to estimate intrinsic vowel duration, but also intrinsic consonant duration. Our demonstration is restricted to English in this article.

Frequency Compression of 7.6 kHz Speech Into 3.3 kHz Bandwidth

Telephone channels restrict the bandwidth of speech signals to approximately 0.3-3.3 kHz, with the consequence that the intelligibility of unvoiced sounds may be significantly impaired. To prevent this band limitation of unvoiced sounds while still confining the speech to the telephonic bandwidth, we propose a scheme which, on recognizing the presence of unvoiced sounds extending to 7.6 kHz, frequency maps them into the band 0.3-3.3 kHz. Four mapping laws are considered and the unvoiced speech is compressed using each law. Frequency demapping is employed, and the law that has the best spectral match to the speech spectrum is selected. Voiced speech is band limited from 0.3 to 3.3 kHz. Results measured over 16 ms, a phoneme, and word durations indicate that the adaptive frequency mapping algorithm significantly enhances the recovered speech compared to telephonic speech. Informal listening experiences support these findings.

Maximum Phoneme Duration of /s/ and /z/ by Children with Vocal Nodules

No AccessLanguage, Speech, and Hearing Services in SchoolsResearch Article1 Jul 1982Maximum Phoneme Duration of /s/ and /z/ by Children with Vocal Nodules Michael P. Rastatter and Melvin Hyman Michael P. Rastatter Google Scholar More articles by this author and Melvin Hyman Google Scholar More articles by this author https://doi.org/10.1044/0161-1461.1303.197 SectionsAboutPDF ToolsAdd to favoritesDownload CitationTrack Citations ShareFacebookTwitterLinked In The maximum phoneme duration (MPD) of /s/ and /z/ task was measured for eight male and eight female prepubertal children with medically diagnosed vocal nodules. The sustained durations for each sound were tallied for both groups of subjects and submitted to appropriate t tests. Mean s/z ratios were compiled also. Results showed that all our subjects' MPD values for both the /s/ and /z/ were shorter than those posited by Tait, Michel, and Carpenter (1980) for their normal subjects. In all cases, the /z/ was sustained longer than the /s/, evidencing a mean MPD ratio of .90. Inspection of the variance in the data showed that all subjects were able to produce the /s/ with greater consistency. Such results indicate the presence of vocal nodules does not affect the MPD s/z ratio, where as the durations of /s/ and /z/ and the variance observed across multiple productions of /z/ are affected and should be considered as diagnostically significant. Additional Resources FiguresReferencesRelatedDetailsCited ByAmerican Journal of Speech-Language Pathology6:1 (65-69)1 Feb 1997The Contributions of Phonatory Volume and Transglottal Airflow to the S/Z RatioMichael D. Trudeau and L. Arick ForrestLanguage, Speech, and Hearing Services in Schools23:2 (163-168)1 Apr 1992The Voiced/Voiceless Phonation Time in Children With and Without Laryngeal PathologyDavid N. Sorensen and Pamela Ann ParkerLanguage, Speech, and Hearing Services in Schools21:1 (60-60)1 Jan 1990Reliability or Difference? A Comment on the Reliability of the S/Z RatioCharles L. Madison and J. Anthony Seikel Volume 13Issue 3July 1982Pages: 197-199 Get Permissions Add to your Mendeley library HistoryReceived: Apr 24, 1980Accepted: Mar 2, 1981 Published in issue: Jul 1, 1982 Metrics Downloaded 18 times Topicsasha-topicsleader-topicsasha-article-typesCopyright & PermissionsCopyright © 1982 American Speech-Language-Hearing AssociationLoading ...

A study of French vowel and consonant durations

To construct a durational model for French speech synthesis-by-rule, 285 words spoken in a frame sentence were analyzed. Durations were measured from digital spectrograms, and were found to vary widely with primary dependence on the following factors: for vowels–height, nasality, and voicing and manner of articulation of ensuing consonants; for consonants–voicing, manner of articulation, and voicing and manner of articulation of adjacent consonants (in consonant clusters); for both vowels and consonants–position of the syllable within the word. A generative model of French phoneme durations in stressed words is presented, and its accuracy as well as the consistency of the durational measurements are examined. Attempts are made to interpret the model along articulatory and phonological lines.

A real‐time phonetic synthesizer

This paper describes a small real‐time speech synthesizer. The synthesizer requires as its input a string of phonemes and the associated duration, pitch, and amplitude parameters. The synthesis scheme uses a large table of stored transitions (dyads) between phonemes. These transitions are stored in terms of LPC‐derived area parameters. To obtain the area parameters for a particular string of phonemes, the appropriate transitions are retrieved and their durations adjusted to reflect the desired durations of phonemes in the string. The speech synthesizer consists of an LS111/23 microcomputer, a digital signal processor chip, and a digital‐to‐analog converter. The microcomputer processes the input, retrieves the stored transitions, adjusts the durations, and smooths the area, pitch, and amplitude parameters across the dyadic boundaries. The smoothed parameters are used to synthesize the speech signal on the digital signal processor.

Maximum duration of sustained /s/ and /z/ in children.

It has been proposed in the clinical literature that a ratio of the maximum duration of /s/ to /z/ may be useful in determining the possible contribution of the respiratory and vocal mechanisms to a disordered voice. The purpose of this study was to obtain preliminary normative data for the maximum phoneme duration (MPD) of /s/ and /z/ and to determine the ratio of /s/ to /z/ in five, seven, and nine year olds. The results indicated that while there were no significant differences in MPD for either /s/ or /z/ between males and females at any one age level, there was a significant increase in the MPD of /s/ and /z/ for both sexes as a function of age. When the ratio of /s/ to /z/ was computed for the subjects, there were no significant differences in these ratios between males and females nor were there significant differences as a function of age.

A durational and spectral analysis of simple French words

Intelligible and natural‐sounding synthetic speech requires accurate specification of phoneme durations and spectra. Toward a goal of French synthetic speech‐by‐rule using a formant synthesizer, a set of 285 words was read in frame sentences by a French Canadian and analyzed via wide‐band spectrograms for durations, formants, and bandwidths. Phoneme durations were found to vary widely with primary dependence on the following factors: for vowels—height, nasality, and voicing and manner of articulation of ensuing consonants; for consonants—voicing, manner of articulation, and voicing and manner of articulation of adjacent consonants (in consonant clusters); for both vowels and consonants—position of syllable within the word and number of syllables in the word. A generative model of French phoneme durations in stressed words will be presented, along with a model for time trajectories of formants and bandwidths in terms of targets and smoothing rules.

Some Notes on Reading

The paper describes a method of estimating a person's reading rate from a small sample of his speech (in other words, a short sentence). Reading rate was obtained by measuring the speaking portion of a considerable amount of reading, and dividing it by the number of phonemes in the material (this is called the grand mean). Out of 108 sentences read by four talkers, 58 having 15 words or less were selected. The ratio of unstressed to stressed syllables, and the number of syllables per occurrence of terminal lengthening were calculated for each of the 58 sentences. Sentences with both of these values in the middle range were considered to have an average phoneme duration close to the grand mean. The prediction was compared to actual readings by four talkers, with 83 percent agreement. The Summary section contains instructions for applying our method of estimating a person's reading rate. Those who are interested should read the Summary section and the Appendix first.

A text-to-speech translation system for Italian

This paper describes a text-to-speech translation system for the Italian language. It is based on a formal translator which converts a graphemic text into the corresponding phonemic representation and on a suprasegmental features descriptor. The translator is implemented according to the formal techniques developed for lexical and syntactical analysis of artificial languages. The suprasegmental features descriptor determines pause positions, phoneme durations and pitch contour on the sole basis of the punctuation marks and the accent marks present in the text.

The effect of selected prosodic errors on the intelligibility of deaf speech

A trained speech pathologist listened to recordings of 40 deaf children, each reading 15 sentences and recorded instances of four types of prosodic errors: adventitious sounds, excessive phoneme duration, pitch breaks, and pauses. The scores were correlated with previously obtained intelligibility scores. It was found that adventitious sounds, except those obviously operating as transitional sounds to ease the movement from one place of articulation to another, had the greatest negative effect on intelligibility, followed by very long duration and pitch breaks. Medium and long pauses, long duration, prolonged closure (i.e., long duration plosives), and adventitious transitional sounds had a lesser negative effect. Short pauses had a small positive effect on intelligibility. The method used to identify the errors was one that can hopefully be evolved into an evaluative instrument for practical use in schools for the deaf.

Acoustic study of juncture perception

Juncture perception as a function of word-initial phonemes was examined in pairs of phrases contrasting in juncture location (e.g., “he praises” and “heap raises”). Contrastive phrases such as this with a wide variety of consonants, consonant clusters, and vowels at the juncture 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. Spectrograms of the phrases were examined for acoustic cues related to the results of the listening test. Juncture perception was good when a word-initial phoneme had a special feature not found in its word-final position, such as aspiration of initial voiceless stops and a glottal stop, laryngealization, breathy voice, or a drop in pitch at the onset of initial vowels. Juncture perception was poor when the same consonant occurred before and after the juncture (e.g., “heap praises”) even when the duration of the doubled consonant was twice as long as the single consonant. The acoustic results suggest that phonemic duration is not an important cue to juncture perception.

Estimate of speaking rate

A technique is described for estimating a person's speaking rate from a short utterance. Four speakers read aloud four stories and also a randomized list of the sentences from those stories. An average time per phoneme was calculated for each sentence from the ratio of total speech time (excluding pauses) to the number of phonemes in the sentence. The longest average duration in these data is 66.4 msec per phoneme; the shortest, 59.6 msec per phoneme. Variation in average duration decreases as sentence duration increases. The story version shows a less variable distribution than the list version. In the story version, all sentences longer than 5 sec have average phoneme durations that fall within ±5 msec of the speaker's overall mean phoneme duration, whereas the list version requires sentences 7 sec long to reach this stability.

Some results from a preliminary study of British English speech rhythm

The paper reports a statistical study that has been carried out in an attempt to gain insight into the nature and underlying causes of rhythm in spoken British English. Two sets of utterances were selected as specially suited to the study, being both widely available as published audio tapes and spoken carefully in an attempt to exemplify normally spoken British English for the benefit of foreign students of the language [Study Units 30 and 39 from M.A.K. Halliday, A Course in Spoken English: Intonation (Oxford University, Oxford, 1970)]. Segmental analyses were performed and the resulting data on segment durations together with additional (higher level) information were prepared for computer analysis. It was found that, although there was some tendency towards isochrony (equally timed rhythmic units), constraints of isochrony accounted for at most 10% of the durational structure of the rhythmic units. Much of the rhythmic structure (about 45%) is adequately accounted for by the mean durations of the constituent phonemes. Other factors play a minor role. [Work supported by the National Research Council of Canada.]

Another consistency in phoneme duration

With natural speech data for vowels and voiceless fricatives, we have shown that the durational behavior of phonemes in fluent speech is very regular when they are classified according to detailed phonological conditions, and that it can be approximated by simple rules [Umeda, J. Acoust. Soc. Am. 58, 434–445 (1975); J. Acoust. Soc. Am. 53, 378(A) (1973)]. Here we report further evidence of consistency in phoneme duration using three groups of consonants that have a closure along the vocal tract—voiceless stops, voiced stops, and nasals—taken from a 20‐min reading by one speaker. Each group has three members: bilabial, dental, and velar. The durational relation between the members, when they do not form clusters with other consonants and when they are classified by stress and boundary conditions, shows a striking similarity across the three groups. The influence of adjacent consonants on the duration of the member is considerable. The variation depends on the relative “consonantalness” of both consonants, and on their manner of excitation and articulation.

Letter: Evidence for a theory of speech production based on observations of the speech of children.

Studies of phoneme duration are reviewed for the purpose of explaining possible mechanisms to account for observed differential durational effects. Information relative to the chronologic appearance of these mechanisms in children is presented showing a developmental sequence and suggesting the possible interrelationship of mechanisms. Findings are related to the model of speech production proposed by MacNeilage and DeClerk.