Prosodic Differences Research Articles

A segmental speech coder based on a concatenative TTS

An extremely low bit rate speech coder based on a recognition/synthesis paradigm is proposed. In our speech coder, the speech signal is produced in a way which is similar to concatenative speech synthesis of text-to-speech (TTS). Hence, database construction, unit selection and prosody modification, which are the major parts of concatenative TTS, are employed to implement the speech coder. The synthesis units are automatically found in a large database using a joint segmentation/classification scheme. Dynamic programming (DP) is applied to unit selection in which two cost functions, an acoustic target cost and a concatenation cost are used to increase naturalness as well as intelligibility. Prosodic differences between the selected unit and the input segment are compensated for by time-scale and pitch modifications which are based on the harmonic plus noise (HNM) model framework. In single speaker tests, the proposed scheme gave intelligible and natural sounding speech at an average bit rate of about 580 b/s.

Prosodic differences in tracheoesophageal speech: Perception and production

Prosody plays an important but often unrecognized role in a listeners interpretation of an utterance. The potential for significant prosodic limitations is often found in speakers who use tracheoesophageal speech due to extirpation of the larynx from cancer. This is because many prosodic features in the normal speaker are controlled by the use of laryngeal muscles [Kent, The Speech Sciences (1997)]. Previous studies showed inconsistencies in intonation and word stress perception and production levels regarding tracheoesophageal speech [Gandour and Weinberg, J. Speech Hear. Res. 26, 142–148; Trudeau et al., J. Commun. Dis. 21, 21–31]. The present study was designed to reevaluate listeners perceptions of tracheoesophageal speech and then describe the production components or limitations leading to these perceptions. Listeners unfamiliar with tracheoesophageal speech listened to normal and tracheoesophageal speakers saying the sentence Put the dishes away as a declarative, query, and exclamation—both in and out of sentential context. Listeners then heard a recording of all speakers and categorized the sentences as to sentence types. Intonation patterns of normal and tracheoesophageal speakers were then compared and contrasted. A more detailed analysis of tracheoesophageal tokens identified as correct nearly all or none of the time was also done.

Acoustic cues to syntatic structure

Duration and fundamental frequency (f0) are used by listeners in on-line sentence processing as cues to syntactic structure [H. Nagel, L. P. Shapiro, and R. Nawy, J. Psycholing. Res. 23, 473–485 (1994)]. Research suggests, however, that speakers may not reliably use duration and f0 to disambiguate syntactic structures [D. W. Albritton, G. McKoon, and R. Ratcliff, J. Exp. Psych. 22, 714–735 (1996)]. The present study explored prosodic differences in a variety of syntactic structures for sentences read by five speakers. Speakers failed to utilize f0 or duration to disambiguate verbs followed by either a direct object or sentential clause; this finding does not support Nagel et al.’s (1994) results for one speaker. In contrast, intransitive verbs were characterized by longer durations and greater f0 excursion as compared to transitive verbs. Acoustic differences for noun phrases (NPs) following transitive and intransitive verbs also were present; NPs following transitive verbs were longer in duration and were associated with a steeper f0 change, as compared to NPs following intransitive verbs. The results suggest that speakers may distinguish some but not all syntactic structures using the prosodic cues of f0 and duration; implications for theories of on-line sentence processing will be discussed. [Work supported by NIH DC00494, DC03347.]

Two-month-old infants' sensitivity to allophonic differences.

The present study investigated 2-month-olds' abilities to discriminate allophonic differences that are potentially useful in segmenting fluent speech. Experiment 1 investigated infants' sensitivity to the kind of distinction that may signal the presence or absence of a word boundary. When tested with the high-amplitude sucking procedure, infants discriminated pairs of items, such as "nitrate" versus "night rate" and "nikrate" versus "nike rate". By greatly reducing the potential contribution of prosodic differences to these contrasts, Experiment 2 evaluated whether the allophonic differences for /t/ and /r/ were sufficient for infants to distinguish the "nitrate" versus "night rate" pair. Infants distinguished "nitrate" from a cross-spliced version of "night rate," which differed only in the allophones for /t/ and /r/ that it included. Thus, infants appear to possess one of the prerequisite capacities (i.e., the ability to discriminate allophonic distinctions) necessary to use allophonic information in segmenting fluent speech.

Imitation of Suprasegmental Patterns by Non-Native Speakers of English

Speakers with limited English proficiency (LEP) usually produce a combination of articulation errors and differences in the prosodic patterns of English. There are abundant measures of articulation, but few objective measures of prosodic performance. The Tennessee Test of Rhythm and Intonation Patterns (TRIP) was selected as a potential measure for distinguishing the prosodic patterns of native and non-native English. First, the TRIP was given to 12 native speakers of Asiatic languages and 12 native speakers of English. The performances of the two groups did not differ significantly, suggesting that the TRIP was not a definitive measure of prosodic differences in English. Second, a group of listeners was asked to identify native versus non-native English speakers based only on selected stimuli from the TRIP and a short sentence of comparable length. The listeners were significantly better able to identify native and non-native speakers when listening to the sentence than when listening to the TRIP items. Clinical application of this information in working with LEP clients is discussed.

The sense of an ending: A case of prosodic delay

Studies of normal and atypical prosodic development show that some children learning English invariably locate the main prosodic prominence at the end of the utterance, even though the main focus of information may come earlier. A case study is presented of David, a speech- and language-impaired child from the West Midlands of England who displayed this prosodic pattern at the age of 5, but not a year later. Since marked prosodic differences exist between the regional accent that he is exposed to and other varieties of English, David's prosodic behaviour is compared to that of adults and children speaking the same West Midlands variety. The analysis draws on techniques developed within conversation analysis to explore the relationship between prosodic detail and interactional behaviour. It is argued that, in order to maintain conversational interaction, children such as David may be using their prosodic resources to delimit unambiguously the end of their turn at talk, and that this is at the expense of cl...

On the analysis of prosody in spontaneous speech with exemplification from Swedish and French

This paper reports on a research project concerning prosody in spontaneous spech. Two questions inaugurate the project. The first one concerns prosodic differences between spontaneous speech and read speech. Evidence from Swedish shows that these differences are not fundamental. The second question concerns the relationship between prosody and discourse categories. A methodology has been developed in order to study this relationship. Four different kinds of analyses are applied: (1) analysis of the discourse structure of the speech corpus without specific reference to prosodic information, (2) auditory analysis in the form of a prosody-oriented transcription, (3) acoustic-phonetic analysis and (4) analysis-by-synthesis. Part of this analysis is illustrated with exemplification from a persuasive monologue in French political rhetoric. Focal accent and contrast in pitch range seem to account for typical prosodic means used during political debate.

The use of prosody in syntactic disambiguation

Prosodic structure and syntactic structure are not identical; neither are they unrelated. Knowing when and how the two correspond could yield better quality speech synthesis, could aid in the disambiguation of competing syntactic hypotheses in speech understanding, and could lead to a more comprehensive view of human speech processing. In a set of experiments involving 35 pairs of phonetically similar sentences representing seven types of structural contrasts, the perceptual evidence shows that some, but not all, of the pairs can be disambiguated on the basis of prosodic differences. The phonological evidence relates the disambiguation primarily to boundary phenomena, although prominences sometimes play a role. Finally, phonetic analyses describing the attributes of these phonological markers indicate the importance of both absolute and relative measures.

Disambiguation of ditropic sentences: acoustic and phonetic cues.

In a previous study, we demonstrated that listeners were highly successful in identifying the intended meaning of spoken ditropic sentences (those which may carry either a literal or an idiomatic meaning) when speakers were instructed to convey the distinction. The present communication reports on acoustic and phonetic analyses carried out with the goal of identifying cues that distinguished the literal and idiomatic utterances. Certain prosodic differences were observed. Literal utterances were systematically longer than idioms. This was partly due to increased use of pauses, as well as to increased duration of major lexical items. Moreover, literal sentences were typically characterized by greater numbers of pitch contours (discernible rise-fall excursions of fundamental frequency) and open junctures than were idiomatic utterances. In addition to suprasegmental contrasts, articulatory distinctions--corresponding to lento-allegro phonological rules--were also observed. These distinctions directly reflect the structural differences intrinsic to the two types of utterances. A literal sentence is formulated by the organization of constituent words and phrases. Idioms, on the other hand, are holistic units, largely nontransparent to syntactic structure or the usual meaning of the lexical members.