Fast Speech Research Articles

The social validity of script training related to the treatment of apraxia of speech

Background: Social validity is an important yet under-examined aspect of treatment efficacy that determines how the effects of treatment are perceived by people other than the clinician/researcher. This is particularly true of treatments to improve the speech of adults with acquired neurogenic disorders. Aims: The purposes of this investigation were to evaluate the social validity of a modified script training treatment protocol, to explore how aspects of a client's speech correspond to varying rater judgements, and to determine which of the listener ratings were most predictive of ratings of overall quality. Methods & Procedures: A total of 124 young, naïve listeners were asked to rate the quality of speech of an 81-year-old woman with moderate–severe apraxia of speech during utterances with varying levels of script correctness (low, medium, high), number of errors (low, high), and rate (slow, faster). Judgements were made on the understandability, ease of production, naturalness, and overall quality of speech. Outcomes & Results: All main effects and interactions were statistically significant. As script correctness, speaking rate, and number of errors increased, listener ratings were significantly more favourable. Interactions demonstrated increasing correctness, error, and rate with significantly more favourable listener ratings. The listener ratings that predicted improved perceptions of overall quality, weighted highest to lowest, were: understandability, naturalness, and ease of production. Error types were analysed and revealed that phrase repetitions appeared to be perceived positively and that unintelligible words and interjections appeared to be perceived negatively. Conclusions: The modified script training protocol applied to a woman with marked apraxia of speech appeared to be socially valid with these naïve raters. Listeners appeared to be sensitive to the amount and the quality of speech output generated by the speaker. Listeners appeared to perceive struggle behaviours as negative, increased speech output (including repetitions and empty speech) as positive, faster speech (closer to a normal speaker's average word per minute) as preferable to slow speech, and more understandable speech and more natural speech as better quality speech. Based on these data, the accuracy and fluency of script production are important treatment goals.

On the intelligibility of fast synthesized speech for people who are blind: A cross-system comparison.

People who are blind increasingly use synthesized speech as a primary output modality when interacting with computers, mobile devices, and web-based services. They typically prefer to listen to the synthesized speech at speeds multiple times real time, and consequently develop strong preferences for particular text-to-speech (TTS) engines and voices. These usage-based preferences lead to potentially incorrect assumptions about which TTS approaches are “best” for fast speech. We report on a cross-system comparison of the intelligibility of fast synthesized speech for users who are blind from birth. We used male and female voices from multiple TTS engines representing the main approaches to TTS: diphone synthesis, unit selection synthesis, HMM-based synthesis, and formant synthesis. We recruited participants from organizations that work with the blind. Each participant listened to and transcribed semantically underspecified sentences from a single TTS engine and voice, spoken at speeds ranging from 300 to 550 words/min. We used TTS engine, algorithm for speeding up synthesized speech, and speaker sex as independent variables, and transcription accuracy as the primary dependent variable. We report differences by TTS approach, speed up method, TTS engine type, and speaker sex.

Evaluation of phonological awareness skills in educable children with down’s syndrome (mental age of 7 to 12 Years) resident in Isfahan, Iran

Introduction: Learning phonological awareness skills plays a very important role in the course of speech and language development and throughout academic education in that these skills help children express their mental concepts more precisely. The aim of the present research was to determine levels of phonological awareness competency among educable children with Down’s syndrome whose mental age ranged between 7 and 12 years and were resident of Isfahan, Iran. Materials and Methods: In this descriptive-analytic study, all Isfahanian children with Down’s syndrome were individually assessed by a phonological awareness test and their scores on different subtests of this test were recorded on a separate sheet exclusively designed for this purpose. Data were statistically analyzed via t-test using SPSS software version 16. Results : Although a superior performance was found in girls, there was no significant difference between boys and girls as far as phonological awareness skills were concerned. In both genders, the highest average score of phonological awareness competency across 12 areas, which were covered by the above-mentioned test, pertained to phonemic synthesis and the lowest average score of such competency was observed in the awareness of middle phoneme. Ranked from simplest to most complex, the following order of competency in phonological skills was observed for both male and female subjects: recognition of the first phoneme, recognition of the last phoneme, awareness of the last phoneme, rhyme detection, alliteration and omission of the first and the middle phoneme. Conclusion: Study results showed that, as a group, phonological awareness skills of girls were superior to those of boys but there were not statistically significant differences between them regarding these skills. This superiority can probably be explained by faster speech and language development in girls and their relatively more advanced use of speech and language skills as it is the case in normal population. Keywords: Down’s syndrome, Phonological awareness, Children

Perception of place‐of‐articulation contrasts in English word‐final consonants by Japanese adult second language learners.

English word‐final consonants are realized differently in connected speech from the same word‐final consonants produced in isolation because of coarticulation with the following segment. Identifying word‐final stops is challenging for Japanese listeners because of the different phonotactic and syllable structures of their L1. This study investigated the perception by Japanese L2 learners and native English controls of word‐final stops /p‐t‐k/, /b‐d‐g/, and /m‐n‐ŋ/ when followed by a word‐initial /p/, /t/, or /k/ in sentences spoken in clear and in fast (casual) speech. Stimuli were minimal triplets differing in place of articulation (e.g., sip, sit, and sick), followed by positively, tauntingly, or cautiously. Participants listened to sentences, e.g., He said the word sit positively (or tauntingly, or cautiously), and identified the target words by choosing one of three written options. Japanese listeners had difficulty identifying English word‐final oral stops in fast speech (26% errors), whereas they did better when the oral stops were released in clear speech (6% errors). Their difficulty identifying word‐final nasals in both conditions (23% errors in clear, 34% errors in fast) reflects a language‐specific perceptual pattern affected by the phonological rules of their L1.

Position and Height Asymmetries in Hiatus Resolution: A case study of Korean VV sequences

In lieu of an abstract, here is a brief excerpt:Typological patterns in synchronic data, for example, the question of why pattern A is more frequent than B across languages, have been one of the most important issues in linguistics, in particular in phonology. Ohala (1993) seeks the answer in human articulatory and/or auditory mechanisms. If a phonetic ‘perturbation’ is not corrected properly in perception, a hypo-correction could occur and it could result in a sound change. In this model, typological patterns are assumed to reflect the very variation in ordinary speech. Two crosslinguistic asymmetries in hiatus reso-lution are the main concerns in this paper. Casali (1996) and Rosenthall (1997) present positional and height asymmetries. They take another approach to the ty-pological patterns in that they suggest universal constraints and rankings to ac-count for the asymmetries in the framework of Optimality Theory (Prince and Smolensky 1993). Crosslinguistically, hiatus resolution such as vowel deletion or gliding is more likely to occur in V1 and high vowels than in V2 and non-high vowels. If these patterns result from human articulatory and auditory mechanisms as Ohala (1993) argues, it would be expected that the production of vowel se-quences will show a pattern of phonetic variation, which is similar to the phono-logical processes. As the first step, a set of acoustic data on Korean hiatus is pre-sented in this paper. The results will show that V1 in hiatus is consistently shorter than V2, which corresponds to the positional asymmetry in Casali (1996). As for the height asymmetry, it will be reported that high vowels are more reduced in fast speech, compared to their durations in slow speech.

Effect of age on clear speech intelligibility.

Previous work has established that elderly adult’s comprehension is more detrimentally affected by noise, multi-talker babble and the absence of contextual cues compared to younger adult’s comprehension. Elderly adult’s abilities to understand fast speech and to detect/recognize brief non-speech and speech sound sequences (e.g., stop gaps) also decline with age. These effects seem to persist even when hearing loss is taken into account. In this study, we explored whether there are parallel effects of age on speech production, i.e., whether elderly adult’s production patterns differ from those of younger adults. Specifically, we examined whether elderly and younger adult’s hyper-articulation clear speech strategies provide similar intelligibility advantage for younger adult listeners. The sentence-in-noise perception results revealed that elderly adult’s clear speech increased intelligibility for young adult listeners. Importantly, the overall elderly adult’s speech intelligibility and the clear speech intelligibility gain were smaller compared to that provided by the speech of younger adults. Acoustic correlates of the differences in the observed clear speech gain provided by older and younger adults are explored. These results suggest that, in addition to perceptual problems, cognitive and hearing changes due to age can affect elderly adult’s speech patterns and intelligibility.

Position and height asymmetries in hiatus resolution: Are they phonetically driven phenomena?

Hiatus is not preferred in many languages and is resolved in various ways. Two typological patterns have been found in hiatus resolution. When hiatus is resolved by a weakening (deletion or gliding) of one vowel, (1) V1 is more likely to be the target than V2 [Casali (1996)] and (2) high vowels are more likely to be the target than non-high vowels [Rosenthall (1997)]. If these patterns are due to human articulatory and/or auditory mechanisms [Ohala (1993)], the patterns would be expected to reflect variations in ordinary speech. In this study Korean CV1V2 nonce words produced by six Korean speakers are analyzed to test whether V1 and high vowels in a realized hiatus are actually weakened in terms of duration. The results showed that V1 is shorter than V2 irrespective of speech rate [F(1,5)=12.90, P<0.02], which was not found in CVCV or CVGV words. It was also found that high vowels are shortened in fast speech to a more degree than non-high vowels [F(1,5)=7.16, P<0.05]. The formant analysis showed that vowel-to-vowel coarticulation effects are greater in CV1V2 than in CVCV, which means that vowels in hiatus are more vulnerable to misperception and consequently, sound change.

Method of speech recognition using variables representing dynamic aspects of speech

A method and computer-readable medium are provided that determine predicted acoustic values for a sequence of hypothesized speech units using modeled articulatory or VTR dynamics values and using the modeled relationship between the articulatory (or VTR) and acoustic values for the same speech events. Under one embodiment, the articulatory (or VTR) dynamics value depends on articulatory dynamics values at pervious time frames and articulation targets. In another embodiment, the articulatory dynamics value depends in part on an acoustic environment value such as noise or distortion. In a third embodiment, a time constant that defines the articulatory dynamics value is trained using a variety of articulation styles. By modeling the articulatory or VTR dynamics value in these manners, hyper-articulated, hypo-articulated, fast, and slow speech can be better recognized and the requirement for the training data can be reduced.

Perceptual learning of time-compressed and natural fast speech

Speakers vary their speech rate considerably during a conversation, and listeners are able to quickly adapt to these variations in speech rate. Adaptation to fast speech rates is usually measured using artificially time-compressed speech. This study examined adaptation to two types of fast speech: artificially time-compressed speech and natural fast speech. Listeners performed a speeded sentence verification task on three series of sentences: normal-speed sentences, time-compressed sentences, and natural fast sentences. Listeners were divided into two groups to evaluate the possibility of transfer of learning between the time-compressed and natural fast conditions. The first group verified the natural fast before the time-compressed sentences, while the second verified the time-compressed before the natural fast sentences. The results showed transfer of learning when the time-compressed sentences preceded the natural fast sentences, but not when natural fast sentences preceded the time-compressed sentences. The results are discussed in the framework of theories on perceptual learning. Second, listeners show adaptation to the natural fast sentences, but performance for this type of fast speech does not improve to the level of time-compressed sentences.

Which words are hard to recognize? Prosodic, lexical, and disfluency factors that increase speech recognition error rates

Despite years of speech recognition research, little is known about which words tend to be misrecognized and why. Previous work has shown that errors increase for infrequent words, short words, and very loud or fast speech, but many other presumed causes of error (e.g., nearby disfluencies, turn-initial words, phonetic neighborhood density) have never been carefully tested. The reasons for the huge differences found in error rates between speakers also remain largely mysterious. Using a mixed-effects regression model, we investigate these and other factors by analyzing the errors of two state-of-the-art recognizers on conversational speech. Words with higher error rates include those with extreme prosodic characteristics, those occurring turn-initially or as discourse markers, and doubly confusable pairs: acoustically similar words that also have similar language model probabilities. Words preceding disfluent interruption points (first repetition tokens and words before fragments) also have higher error rates. Finally, even after accounting for other factors, speaker differences cause enormous variance in error rates, suggesting that speaker error rate variance is not fully explained by differences in word choice, fluency, or prosodic characteristics. We also propose that doubly confusable pairs, rather than high neighborhood density, may better explain phonetic neighborhood errors in human speech processing.

A Duration-Dependent Account of Coarticulation for Hyper- and Hypoarticulation

Previous studies investigating anticipatory coarticulation in emphatically stressed CV sequences and during fast speaking rates reported that three factors contributed to the overall extent of the documented coarticulation. These factors were: (1) vowel identity, (2) vowel space expansion (emphasis) or reduction (fast rate), and a hypothesized (3) ‘deeper’ and ‘shallower’ stop closure contact in emphatic and faster speech, respectively. The objective of the current research was to conceptually and quantitatively unify these two studies. This was accomplished by showing that the opposite changes to frequency onsets of F2 transitions due to emphatic and rapid speech systematically vary as a function of the durational changes in the stop closure interval. Specifically, the decrease in coarticulation in emphatic speech is characterized by increases in F2 onsets and longer stop closures (relative to a normal baseline); the increase in coarticulation due to rapid speech shows concomitant decreases in F2 onsets coinciding with shorter stop closure intervals. Vocal tract area function simulations corresponding to emphatic and reduced speech implicitly support ‘deeper’ and ‘shallower’ closure contacts as a third factor contributing to the overall extent of anticipatory CV coarticulation.

Comprehension of Speeded Discourse by Younger and Older Listeners

Researchers have argued that older adults are more adversely affected by speeding speech than are younger adults. However, the age effects usually occur when (1) the speech materials are artificially speeded to rates well above those that occur in natural speech; (2) the speeding method introduces distortions that tax the older adult's auditory processes; and (3) the speech materials are simple sentences or very short passages. This study evaluated whether older adults are disadvantaged when listening to extended discourse (10- to 15-min lectures) speeded to a rate near to the limit of normally encountered fast speech (240 words/min) with a minimum of acoustic distortion. Perceptual difficulty was further manipulated by presenting stimuli in either quiet or with a 12-talker background babble. Younger and older adults had more difficulty recalling the details of the discourse and integrating their contexts when stimuli were presented at faster rates and in higher levels of background noise. Although each of these manipulations were found to cause large differences in performance, the age groups were generally found to perform analogously in most conditions. Potentially the availability of semantically rich materials, and the extended durations of the passages, allowed the older adults an opportunity to adjust to the faster speech rates and maintain performance levels similar to younger adults.

Processing of fast speech by elderly listeners

This study investigates the relative contributions of auditory and cognitive factors to the common finding that an increase in speech rate affects elderly listeners more than young listeners. Since a direct relation between non-auditory factors, such as age-related cognitive slowing, and fast speech performance has been difficult to demonstrate, the present study took an on-line, rather than off-line, approach and focused on processing time. Elderly and young listeners were presented with speech at two rates of time compression and were asked to detect pre-assigned target words as quickly as possible. A number of auditory and cognitive measures were entered in a statistical model as predictors of elderly participants' fast speech performance: hearing acuity, an information processing rate measure, and two measures of reading speed. The results showed that hearing loss played a primary role in explaining elderly listeners' increased difficulty with fast speech. However, non-auditory factors such as reading speed and the extent to which participants were affected by increased rate of presentation in a visual analog of the listening experiment also predicted fast speech performance differences among the elderly participants. These on-line results confirm that slowed information processing is indeed part of elderly listeners' problem keeping up with fast language.

Rate conditioned variability in Western Andalusian Spanish aspiration.

Western Andalusian Spanish normally shows aspiration of /s/ in coda position, but post-, rather than pre-, closure aspiration in /s/ + /p,t,k/ sequences. This has been hypothesized to be the result of gestural reorganization, with pre-closure aspiration the result of anti-phase coordination and post-closure the result of in-phase [Torreira, 9th HLS, 2007]. This experiment explores this hypothesis by leveraging the fact that fast speech can lead to instability in gestural organization and even a spontaneous switch from anti- to in-phase coordination [Turvey, Am. Psych., 1990]. Subjects produced words with the target /s/+/p,t,k/ sequences in time with an acoustic metronome that steadily increased in rate. Results show three patterns of production: consistent preaspiration, consistent postaspiration, and a transition from pre- to post-aspiration as rate increased, corresponding to anti-phase, in-phase, and an anti- to in-phase transition, respectively. Excluding the trials with consistent postaspiration, which would not be expected to transition given their stable in-phase coordination, postaspiration increases as rate increases. If there were no change in gestural coupling, we would expect the opposite pattern: as rate increases, VOT should decrease. The result thus supports the original hypothesis that postaspiration in these sequences is the result of gestural re-phasing.

Lexical tone and segmental coarticulation.

In studying anticipatory coarticulation in emphatic CV sequences [Lindblom et al., J. Acoust. Soc. Am. 121, 3802–3813 (2007)] and in fast speech [Agwuele et al., Phonetica 68, 194–209 (2008)], it was shown that prosodically induced modulations of speech introduce additional sources of phonetic variation separate from those generated by vowel context. Both studies were successful in dissociating these additional prosodic effects from the underlying contextual vowel variation and quantifying them. The focus of this current investigation was to study CV coarticulation, using a similar set of regression metrics, but now applied to a tonal language (Yoruba). It is to be determined whether lexical tone impacts the coarticulation of consonant and vowel. Specifically, words, do tones, produce additional effects on anticipatory CV coarticulation separate from those generated by vowel‐context similar to emphatic stress and increased tempo as documented by the aforementioned studies? Finally, of interest also is to examine whether different lexical tones impact C+V coarticulation differently.

DECtalk™ and VeriVox™: Intelligibility, Likeability, and Rate Preference Differences for Four Listener Groups

The intelligibility, likeability, and speech presentation rates for two types of popular synthesized speech, DECtalk™ and VeriVox™, were investigated across four age groups ranging from 6–85 years old. Results for all age groups revealed differences in intelligibility between synthesized adult and child voices, with intelligibility scores for child synthesized voices significantly below those for adult synthesized voices. The most popular synthesized voices were from the VeriVox family. Of interest is that preferred speech presentation rate, expressed as mean words per minute, increased as a function of age, with the oldest listeners preferring the fastest speech presentation rate. This is inconsistent with previous studies that have investigated rate preferences among older adults, and could reflect the different synthesized speech options employed in this study. Implications of these findings for the improvement of text-to-speech products and the practice of AAC are discussed, along with suggestions for further research.

Moving to the Speed of Sound: Context Modulation of the Effect of Acoustic Properties of Speech

Suprasegmental acoustic patterns in speech can convey meaningful information and affect listeners' interpretation in various ways, including through systematic analog mapping of message-relevant information onto prosody. We examined whether the effect of analog acoustic variation is governed by the acoustic properties themselves. For example, fast speech may always prime the concept of speed or a faster response. Alternatively, the effect may be modulated by the context-dependent interpretation of those properties; the effect of rate may depend on how listeners construe its meaning in the immediate linguistic or communicative context. In two experiments, participants read short scenarios that implied, or did not imply, urgency. Scenarios were followed by recorded instructions, spoken at varying rates. The results show that speech rate had an effect on listeners' response speed; however, this effect was modulated by discourse context. Speech rate affected response speed following contexts that emphasized speed, but not without such contextual information.

Fast speech recognition to access a very large list of items on embedded devices

In this paper, we propose a fast decoding algorithm to recognize a very large number of item names on a resource-limited embedded device. The proposed algorithm is based on a multi-pass search scheme. The algorithm is composed of a two-stage HMM-based coarse match and a detailed match. The two-stage HMM-based coarse match is aimed at rapidly selecting a small set of candidates that are assumed to contain a correct hypothesis with high probability, and the detailed match re-ranks the candidates by performing acoustic rescoring. The proposed algorithm is implemented on an in-car navigation system with a 32-bit fixed-point processor operating at 620 MHz. The experimental result shows that the proposed method runs at maximum speed 1. 74 times real-time on the embedded device while minimizing the degradation of the recognition accuracy for a 220 K Korean Point-of-Interest (POI) recognition domain.

Children with a cochlear implant: Characteristics and determinants of speech recognition, speech-recognition growth rate, and speech production

The objectives of the study were to describe the characteristics of the first 79 prelingually deaf cochlear implant users in Norway and to investigate to what degree the variation in speech recognition, speech- recognition growth rate, and speech production could be explained by the characteristics of the child, the cochlear implant, the family, and the educational setting. Data gathered longitudinally were analysed using descriptive statistics, multiple regression, and growth-curve analysis. The results show that more than 50% of the variation could be explained by these characteristics. Daily user-time, non-verbal intelligence, mode of communication, length of CI experience, and educational placement had the highest effect on the outcome. The results also indicate that children educated in a bilingual approach to education have better speech perception and faster speech perception growth rate with increased focus on spoken language.

Processing of Japanese vowel devoicing and the effect of speech rate

Vowel devoicing (or reduction) in Japanese is an interface between the phonetic and phonological levels (Vance, 1987; Yoshida, 2002). The phonological rule for vowel reduction defines that high vowels (/i, u/) undergo reduction when between voiceless consonants or between a voiceless consonant and a pause. However, there are some cases of vowel reduction that do not fit the rule. A high vowel adjacent to a voiced consonant sometimes becomes reduced, especially in casual speech (Arai, 1999). Even a nonhigh vowel can undergo reduction (Arai, 1999; Vance, 2004). In this study, the effect of speech rate on the processing of reduced vowels is examined, particularly for vowels that are only likely to be reduced in casual, fast speech. How do listeners recognize vowels that are reduced in fast speech, or the words containing them? A lexical decision experiment showed that listeners found it easier to process fast speech containing reduced vowels than careful speech of isolated words containing them. They also found vowel reduction acceptable for nonhigh vowels in fast speech. This suggests that listeners use knowledge of both phonology and speech rate variability in processing connected speech.