Interword Pauses Research Articles

Adaptative and non-adaptive phonetic variation: Clear speech and second-language speech

Two independent lines of research have revealed a somewhat surprising convergence between clear speech and second-language (L2) speech. Specifically, relative to conversational L1 speech, both clear speech and L2 speech typically exhibit slower speaking rates (fewer syllables per second), longer and more frequent inter-word pauses, and less segment- and syllable-level reduction. These phonetic features result in a general pattern of lower information density (more phonetic material produced for a given linguistic message or text) compounded by longer utterance durations to yield substantially lower information transmission rates for both clear speech and L2 speech relative to conversational L1 speech. However, clear speech and L2 speech diverge in terms of intelligibility. While clear speech is a highly effective strategy for enhancing overall intelligibility, L2 speech is typically characterized by greater likelihood of listener errors in word recognition accuracy. This pattern of convergent phonetics and divergent intelligibility highlights the complicated relationship between phonetic form and communicative efficiency. While clear speech is a listener-oriented talker adaptation to listener limitations, L2 speech reflects talker-oriented talker-listener misalignment. The phonetic similarities between these adaptative (clear speech) and non-adaptive (L2 speech) sources of variation underscore the dynamics of human speech communication with the talker-listener relation as the fulcrum.

WavBERT: Exploiting Semantic and Non-semantic Speech using Wav2vec and BERT for Dementia Detection.

In this paper, we exploit semantic and non-semantic information from patient's speech data using Wav2vec and Bidirectional Encoder Representations from Transformers (BERT) for dementia detection. We first propose a basic WavBERT model by extracting semantic information from speech data using Wav2vec, and analyzing the semantic information using BERT for dementia detection. While the basic model discards the non-semantic information, we propose extended WavBERT models that convert the output of Wav2vec to the input to BERT for preserving the non-semantic information in dementia detection. Specifically, we determine the locations and lengths of inter-word pauses using the number of blank tokens from Wav2vec where the threshold for setting the pauses is automatically generated via BERT. We further design a pre-trained embedding conversion network that converts the output embedding of Wav2vec to the input embedding of BERT, enabling the fine-tuning of WavBERT with non-semantic information. Our evaluation results using the ADReSSo dataset showed that the WavBERT models achieved the highest accuracy of 83.1% in the classification task, the lowest Root-Mean-Square Error (RMSE) score of 4.44 in the regression task, and a mean F1 of 70.91% in the progression task. We confirmed the effectiveness of WavBERT models exploiting both semantic and non-semantic speech.

Distinguishing Spastic and Dystonic Features in a Patient With a History of Brainstem Encephalitis: A Clinical Case Study

Purpose One role of a speech-language pathologist (SLP) is to differentially diagnose motor speech disorders to aid physicians, typically neurologists, in establishing a medical diagnosis. The present case study describes the clinical problem solving undertaken by the SLP to characterize the nature of neurological deficits to develop an appropriate management plan. Method The patient is a 29-year-old man with a history of brainstem encephalitis who presented for evaluation of dysarthria, dysphagia, and gait difficulties. The neurological exam characterized a spastic dysarthria. The SLP conducted a motor speech examination expressly for the purpose of differential diagnosis. Results The patient exhibited features of spastic dysarthria including strained phonation and hypernasality, equivocally slow rate associated with intermittent interword pauses rather than lengthened segments, and articulatory imprecision in the absence of weakness or slowness of the articulators. Defining features of spastic dysarthria absent from his speech were monopitch and monoloudness, as well as slow alternate motion rates. A diagnosis of hyperkinetic dysarthria associated with dystonia was informed by (a) hypernasality disproportionate to other deviant speech features, with preserved reflexive velar movements and nonspeech movements; (b) adventitious lip pursing; (c) modest improvement in speech with sensory tricks; (d) mild and intermittent phonatory strain; and (e) normal nonspeech oral motor function. Conclusions This case demonstrates the role speech-language pathology can play in the neurological assessment and management process. The recognition by the SLP of hyperkinetic dysarthria led to additional diagnostic tests and to a broader range of options for medical management.

Internalising symptoms and verbal working memory in school-age children: A processing efficiency analysis.

The current study investigated differential contributions of internalising symptoms (state anxiety, trait anxiety, depression) to school-age children's verbal short-term (STM) and working memory (WM) span accuracy and efficiency (microanalysis of response times). Children's (N = 125, Mage = 11.44 years) STM/WM was assessed with simple/complex span tasks. Our analyses revealed that: (a) children with high levels of state anxiety displayed reduced simple span accuracy (on Word span) and poorer efficiency on both simple (preparatory intervals, interword pauses) and complex span (preparatory intervals) response time segments; (b) trait anxiety was a negative predictor of children's complex span accuracy, as well as their efficiency on both simple (word durations) and complex span (interword pauses) response time measures; (3) depressive symptoms predicted longer simple span interword pauses. Findings indicate that while all internalising symptoms were predictive of children's poorer memory search efficiency, especially during the "silent", executive intervals (interword pauses), anxiety symptoms were specifically predictive of children's impaired span accuracy and other efficiency indicators (preparatory intervals, word durations). The study highlights the differential contributions of state, trait anxiety, and depressive symptoms to STM/WM in children, emphasising the need to measure both accuracy and efficiency to assess the role that such symptoms play in children's performance.

Effects of inter-word pauses on speech intelligibility under long-path echo conditions

Long-path echo is a salient factor that causes the degradation of the intelligibility of speech transmitted through a wide area outdoor environment or a very large indoor space using public-address systems. To robustly transmit speech information under such conditions, it is important to overcome this effect by controlling the characteristics of speech sounds. In this study, we consider the effects of inserting pauses between the words of a sentence. We performed word intelligibility tests using a series of four continuous words, called a quadruplet. Various pause lengths and long-path echo patterns were applied to the quadruplet. The results of the experiments demonstrate that word intelligibility under a long-path echo is significantly improved by the insertion of pauses between the words. Intelligibility can approach the same levels observed in the absence of echoes for a pause length of approximately 200 ms, which is almost the same as the length of 1-mora for the words used in the experiments. Moreover, this 200 ms pause is known to be sufficient to improve speech recognition in older adults. These results suggest that inter-word pauses of a length of approximately 1-mora can generally enhance the robustness of speech communication systems when used under a severe environment.

Speech intelligibility, speaking rate, and vowel formant characteristics in Mandarin-speaking children with cochlear implant

The effects of the use of cochlear implant (CI) on speech intelligibility, speaking rate, and vowel formant characteristics and the relationships between speech intelligibility, speaking rate, and vowel formant characteristics for children are clinically important. The purposes of this study were to report on the comparisons for speaking rate and vowel space area, and their relationship with speech intelligibility, between 24 Mandarin-speaking children with CI and 24 age-sex-education level matched normal hearing (NH) controls. Participants were audio recorded as they read a designed Mandarin intelligibility test, repeated prolongation of each of the three point vowels /i/, /a/, and /u/ five times, and repeated each of three sentences carrying one point vowel five times. Compared to the NH group, the CI group exhibited: (1) mild-to-moderate speech intelligibility impairment; (2) significantly reduced speaking rate mainly due to significantly longer inter-word pauses and larger pause proportion; and (3) significantly less vowel reduction in the horizontal dimension in sustained vowel phonation. The limitations of speech intelligibility development in children after cochlear implantation were related to atypical patterns and to a smaller degree in vowel reduction and slower speaking rate resulting from less efficient articulatory movement transition.

The differential effects of dysarthria type on lexical segmentation.

Although intelligibility deficits are common across various forms of dysarthria, it was hypothesized that the causes of intelligibility reductions depend on the ways the degraded signal challenges listener cognitive-perceptual strategies. This hypothesis was confirmed by analyzing the lexical boundary errors (LBE) elicited by 44 speakers with one of four forms of dysarthria (ataxic, hypokinetic, hyperkinetic, mixed spastic-flaccid). Error patterns from 60 listener transcripts revealed reliance on metrical stress to segment the 80 six-syllable phrases. However, reductions and disturbances in the acoustic cues to syllabic stress associated with each form of dysarthria resulted in different LBE patterns. Explanations for these unique patterns are found, in part, in the metrical structures of the dysarthric speech. For example, the temporal relationships among stressed and unstressed syllables varied widely in hyperkinetic speech, which ostensibly prevented listeners from optimally exploiting metrical stress as a cue for segmentation. The equal-even temporal structure of the spastic-flaccid speech resulted in the highest proportion of boundary insertions. The presence of interword pauses facilitated the assignment of word onset to strong syllables. Results will be presented relative to models of lexical segmentation and clinical implications for targets of remediation in the dysarthrias. [Work supported from NIH-NIDCD 5R01 DC6859.]

The endurance of children’s working memory: A recall time analysis

We analyze the timing of recall as a source of information about children’s performance in complex working memory tasks. A group of 8-year-olds performed a traditional operation span task in which sequence length increased across trials and an operation period task in which processing requirements were extended across trials of constant sequence length. Interword pauses were longer than are commonly found in immediate serial recall tasks yet shorter than for reading span. These pauses increased with the demands of recall, decreased across the output sequence, and were to some extent predictive of scholastic ability. Overall, timing data illustrate that recall in working memory tasks involves subtle processes of item access rather than simple readout of information from an immediate store.

The effects of four variables on the intelligibility of synthesized sentences

The experiments reported here examined the effects of four variables on the intelligibilty of synthetic speech: (1) listener age, (2) listener experience, (3) speech rate, and (4) the presence versus absence of interword pauses. The stimuli, eighty IEEE–Harvard Sentences, were generated by a DynaVox augmentative/alternative communication device equipped with a DECtalk synthesizer. The sentences were presented to four groups of 12 listeners each (children (9–11 years), teens (14–16 years), young adults (20–25 years), and adults (38–45 years). In the first experiment the sentences were heard at four rates: 105, 135, 165, and 195 wpm; in the second experiment half of the sentences (presented at two rates: 135 and 165 wpm), contained 250 ms interword pauses. Conditions in both experiments were counterbalanced and no sentence was presented twice. Results indicated a consistent decrease in error rates with increased exposure to the synthesized speech for all age groups. Error rates also varied inversely with listener age. Effects of rate variation were inconsistent across listener groups and between experiments. The presences versus absences of pauses affected listener groups differently: The youngest listeners had higher error rates, and the older listeners lower error rates when interword pauses were included in the stimuli. [Work supported by St. John’s University and New York City Board of Education, Technology Solutions, District 75.]

The effects of speech rate and listening experience on the intelligibility of synthesized speech

Some persons with severe communication disorders are only able to communicate effectively using speech synthesizers, although even the most intelligible synthesis systems (e.g., DynaVox-DECTalk) have limitations. Listener success has been reported to vary with listener age, listener experience with synthesized speech, speech rate, and the presence of pauses between words. This series of experiments has examined the effects of listener age and experience, and speech rate and interword pauses on the perception of synthesized speech. Eighty IEEE-Harvard Sentences were presented to four groups of 12 listeners each [children (9–11 years), teens (14–16 years), young adults (20–25 years), and adults (38–45 years)], in two experiments. In the first experiment (40 sentences) speech rate was varied (105, 135, 165, and 195 wpm); in the second experiment (40 sentences), 7–12 days later, interword pauses occurred in half the sentences. Conditions were counterbalanced; no sentence was presented twice. Preliminary results indicate little effect of speech rate but a substantial effect of experience for the children: they improved substantially through experiment 1; the improvement carried over into experiment 2. Parallel results will be presented for all age groups. [Work supported by St. Johns University and New York City Board of Education, Technology Solutions, District 75.]

The Differential Maturation of Two Processing Rates Related to Digit Span

Recent studies have proposed that rapid-speaking durations predict short-term memory spans. However, N. Cowan et al. (1998,Journal of Experimental Psychology: General,127, 141–160) foundtwoseparate types of processing durations related to digit span for children in Grade 1 (7–8 years), Grade 3 (9–10 years), and Grade 5 (11–12 years): rapid-speaking durations (thought to index the rate of covert articulation) and the durations of interword pauses in the memory span task responses (thought to index the rate of short-term memory retrieval). The present analysis additionally establishes the differential maturation of those two durations. Within-age correlations between span and rapid-speaking durations were significant only in first graders; correlations between span and interword pauses, only in fifth graders. When subsamples were selected to match spans across age groups, both types of duration also were matched between Grades 1 and 3. However, fifth graders had considerably shorter interword pauses than their third-grade counterparts. Thus, a particular memory span is accompanied by different profiles of processing rates in children of different ages.

Modularity of Prosody: Autonomy of Phonological Quantity and Intonation in Aphasia

The Finnish language is used here to enrich our conception of thedysprosodyhypothesis of the output in Broca's agrammatism, i.e., of the claim that agrammatic speech is characterized by aberrant timing and intonation patterns. The native language of the two agrammatic aphasics of the present study places a high functional load on quantity (acoustically: timing/duration) in its phonology, as the short–long quantity opposition concerns both vowels and consonants. Moreover, the quantity opposition in Finnish is not local, i.e., syllable-internal as in many Germanic languages, because the phonetic quantity values are determined by a disyllabic sequence. In contrast to the high functional load of quantity in its lexical phonology, Finnish makes little, if any, grammatical use of intonation. In the present acoustic analyses it was noted that the two Broca's aphasics have preserved the disyllabically determined quantity oppositions, although the speech of the aphasics is characterized by long interword pauses, i.e., by syntactic dysprosody. In contrast to the asyntactic timing patterns, the declination line of intonation is retained in the output of the present aphasics in spite of the low grammatical value of intonation in the language. The present results further dispute the characterization of agrammatic speech output as being dysprosodic, or more specifically atemporal, as lexical quantity and syntactic tonal patterns can be retained in a speech mode that is replete with long interword pauses. More generally, the observations will corroborate the autonomy (modularity) of lexical and syntactic processes in oral language production.

A two-step segmentation method for automatic recognition of speech of persons who are deaf.

The development and use of a two-step word segmentation method for automatic recognition of the speech produced by deaf persons are described in this paper. This method incorporates the segmental and temporal characteristics of deaf speech, particularly the intra- and inter-word pauses, to achieve accurate recognition. In the first step, estimates of word boundaries were obtained by detecting these pauses. In the second step, the word boundaries were determined by matching all possible groupings of speech segments against all reference templates during recognition. This method was applied to recognition of isolated words and connected-speech utterances produced by two deaf speakers. Recognition rates of 93.01 percent and 81.81 percent were obtained in recognizing isolated words and connected speech, respectively.

Adults' word recall and word repetition

This experiment investigated age group differences in working memory by examining the effects of word length on adults' recall span and repetition rate. College students and adults, 60 to 94 years of age, recalled lists of one-, two-, or three-syllable words and repeated aloud pairs of one-, two- or three-syllable words. Word recall spans and word repetition rates were computed. Main effects of age group and word length were obtained on both measures, although the interactions were not significant. A second analysis examined the relationship between individuals' recall span and their repetition rate. Across all age groups combined, recall span was a linear function of repetition rate and accounted for 25 percent of the variance on the recall task. A reanalysis of the word repetitions revealed that older adults' word durations and inter-word pauses are longer than young adults'.

Temporal characteristics in the fluent speech of child stutterers and nonstutterers

Past research reveals that adult stutterers often possess aberrant speech-timing abilities. As yet, only Healey and Adams (1981) have investigated these abilities in child stutterers. The present study attempted to assess the speech-timing skills of child stutterers and normal-speaking children while performing both simple and complex speech. Each subject was required to participate in two tasks: a sentence repetition task for simple speech and a story-retelling task for complex speech. Results supported those found by Healey and Adams, indicating that child stutterers have speech-timing abilities similar to normal-speaking children in simple speech tasks. The complex task, however, revealed that stutterers exhibited more frequent and longer interword pauses than nonstutterers. Two hypotheses generated from past research are offered to explain these differences. A third untested theory is also presented.

Automatic prototype selection for continuous speech recognition

Automatic prototype selection of phones, and other entities allows large amounts of data to be processed very efficiently. The HEAR acoustic processor, wholly developed and tested for continuous speech recognition at the IBM Thomas J. Watson Research Laboratory in Yorktown Heights, NY, applies Baum's algorithm for prototype selection. Peformance on two tasks with extensive training on a single speaker, in a quiet, high quality recording environment are reported for HEAR, used in conjunction with the standard training and decoding programs developed by the IBM Research Continuous Speech Recognition Group. Sentence recognition on a set of 125 test sentences (1010 words) of the artificial language 250 word “New Raleigh Language” is 100%. The word recognition rate on a set of ten test sentences (282 words) of the natural language, 1000 word “Laser‐1000 Task” is 80% and reflects the performance of all the system components; e.g., search strategy, language model, pruning procedures, acoustic processor, etc. For only 3.2% of the words is the correct word hypothesized, fully matched, and rejected in favor of an incorrect word. These errors are comprised exclusively of substitutions and deletions of short function words (e.g., “of,” “the,” etc.), 2.2%, and deleted commas (realized acoustically only by optional interword pauses), 1.0%.

Control of speech prosody in Broca's aphasia

A typical clinical description of the speech of Broca's aphasics includes the characteristics of nonfluency, agrammatism, and dysprosody. This study aims to objectively evaluate the subjective impression of dysprosody using computerized techniques to measure fundamental frequency (F0) and segmental timing. We examined three rudimentary aspects of normal speech production: (1) terminal falling F0 contour in utterance-final position, (2) declination of F0 peaks throughout the course of an utterance, and (3) segmental lengthening in utterance-final position. Seventy-five two-word utterances, extracted from the spontaneous speech of three Broca's aphasics, were found to exhibit both terminal falling F0 contours and declination despite interword pauses of up to 3.5 s. However, the timing of these utterances showed no utterance-final lengthening. These results are discussed in terms of the speech planning capabilities of these patients and the apparent dissociation between programming of F0 and timing. A controlled experiment was conducted in order to rule out the possibility that a phonetic influence masked utterance-final lengthening in the spontaneous speech sample; the results confirmed the previous observations for both segmental timing and F0. [Supported by NIH Grants NS 11408, NS 13028, and the MIT Research Laboratory of Electronics.]

Stimulus factors and listening strategies in auditory memory: A theoretical analysis

A theory for the temporal course of perceptual coding during auditory short-term memory tasks is described. A low-level “sensor” and a high-level “identifier” process item and order information from stimulus word-strings. Both word durations and interword pauses are needed for this processing, and if adequate processing time is unavailable, perceptual delays will lead to degraded recall performance. Stimulus parameters and subject strategies interact to determine the quality and the order in which stimulus items are coded. These hypotheses are supported both by experiments using grammatical sentences and those using lists of unrelated words.