Search

What makes a colored image, e.g. an abstract painting or a landscape, look pleasing? We hypothesized that a preference for complex color compositions, such as paintings and images of natural scenes, might be related to how natural the colors are perceived. We tested this possibility with two experiments in which the degree of naturalness of images was manipulated by rotating their color gamut rigidly in the color space CIELAB. This changed just the hue composition, but preserved saturation and lightness. In the first experiment we obtained individual scaling curves for perceived naturalness and for preference as a function of the angle of gamut rotation for a small set of images. The naturalness and preference scaling curves were found to be largely similar and their maxima were close to the original image. In the second experiment, we tested whether this effect generalized to a larger set of images. We used a simultaneous 5AFC procedure where in each trial participants had to select the most natural or the most preferred image from five different rotations of the color gamut. The results confirmed the first experiment and showed that, in general, the images perceived as the more natural tend to be the ones that are preferred. Together these results show that perceived naturalness and preference are indeed perceptually closely related and may be driven by related mechanisms. [ABSTRACT FROM AUTHOR]

Highlights • Randomly constructed 1+1 NN compounds receive an overall low score. • Elasticity has a significant effect on acceptability score. • Naturalness and frequency have a significant effect on acceptability score. Abstract Chinese 1+1 NNs refer to compounds made of two monosyllabic nouns, such as shū - bāo 'book-bag' and jī-dàn 'chicken-egg'. They occur frequently, do not violate any prosodic constraint, and appear to be highly productive. However, in Qin and Duanmu (2017) , it was found that the mean acceptability ranking was quite low for randomly constructed 1+1 NNs. Since the focus of Qin and Duanmu (2017) was on a different issue, the set of 1+1 NNs was small and not of balanced design. In this study, we gathered acceptability scores from 17 Chinese native speakers on 1000 randomly constructed 1+1 NNs. It is found that the mean acceptability score is only 22.6%, even lower than that of Qin and Duanmu (2017). In addition, while Qin and Duanmu (2017) found ambiguity, naturalness, and frequency to have significant effects on acceptability scores, the present study did not find ambiguity to be relevant. We also found that 1+1 NNs consisting of two elastic nouns are significantly less acceptable than those consisting of one or two non-elastic nouns, although the effect size is rather small. Possible reasons for the low acceptability of 1+1 NN compounds are discussed, including semantic transparency, grammatical relations between the two nouns of a compound, and modern vs. classic vocabulary. [ABSTRACT FROM AUTHOR]

Abstract This paper presents a talking gesture generation system based on Generative Adversarial Networks, along with an evaluation of its adequateness. The talking gesture generation system produces a sequence of joint positions of the robot's upper body which keeps in step with an uttered sentence. The suitability of the approach is demonstrated with a real robot. Besides, the motion generation method is compared with other (non-deep) generative approaches. A two-step comparison is made. On the one hand, a statistical analysis is performed over movements generated by each approach by means of Principal Coordinate Analysis. On the other hand, the robot motion adequateness is measured by calculating the end effectors' jerk, path lengths and 3D space coverage. Highlights • Social robots need to show natural body language when talking. • A generative approach should allow to create novel movements. • GANs are regarded as producing better samples than other methods. • Statistical analysis confirms that GANs are better than other approaches. • Videos show the naturalness of the robot behavior. [ABSTRACT FROM AUTHOR]

Highlights • Close temporal proximity between focused words and gestural apices was found. • Unlike previous findings, apex alignment was not dependent on F0 in Cantonese. • Both consistency and individuality in apex alignment pattern was observed. • Most speakers consistently aligned apices to the same syllable of focused words. • They differed in the syllable aligned while all maintained gestural naturalness. Abstract This study investigates the temporal relationship between focus prosody and co-speech pointing gestures in Hong Kong Cantonese. Previous studies have generally shown a close temporal proximity between prosodic and gestural prominence: Gestural prominence tends to be aligned with stressed syllables or words. However, this finding was based solely on studies of stress and pitch-accent languages, and no study has yet tested the phenomenon in a non-stress tone language. Ten native speakers of Hong Kong Cantonese participated in a picture-verification task in which pointing was elicited along with verbal corrections. The acoustic results showed that the corrective focus was marked solely by an on-focus durational increase. The gestural results revealed that there was an alignment between prosodic and gestural prominence, as most of the gesture apices were produced within the focused words. However, in contrast to previous findings, no significant effect of F0 (tone) or focus position was found. Instead, most speakers consistently aligned their apices with the same syllable position in disyllables. Based on the current findings, the prosodic anchor of prosody-gesture alignment is suggested to be the focused word in this language. [ABSTRACT FROM AUTHOR]

This article investigates the interpretative properties of two clefts in French; the well-known c’est -cleft and the under-studied y’a -cleft. A prevalent assumption is that, when they signal narrow-focus, these two clefts differ with respect to exhaustivity ; the former specifies a unique referent for the focus variable, but not the latter. Empirical evidence from a forced-choice task suggests that this analysis is going down the right path. Yet, the paper argues for a refined understanding of the conditions of use for these two clefts, positing that c’est - and y’a -clefts do not occur in the exact same narrow-focus context. Rather, their alternation is linked to the type of question asked—a feature absent from past studies. In a nutshell, I argue that c’est -clefts are more naturally found with questions that require a full answer (i.e. Mention-All questions), whereas y’a -clefts are more appropriate in contexts where the question allows for a non-maximal answer (i.e. Mention-Some questions). Additionally, when y’a -clefts occur as answers to Mention-All questions, they convey an ignorance inference, which is much weaker in Mention-Some contexts. Evidence supporting this proposal comes from a second experiment; a rating task whose results also suggest that the type of question asked not only influences naturalness, but also the strength of the inferences respectively associated with the two clefts (exhaustivity and ignorance). [ABSTRACT FROM AUTHOR]

Some seemingly robust linguistic generalizations remain controversial for years, often for lack of experimental verification. As a case study, we examine word length preferences in Chinese NN compounds. Since the original observation of Lü (1963) , there has been a broad consensus that 1+2 (monosyllabic + disyllabic) is ill formed, while 2+1, 2+2 and 1+1 are well-formed. In addition, it has been shown that the preferences can be derived from metrical principles. However, a judgment experiment is yet to be offered to verify the preferences, and without it some scholars remain skeptical. We offer such an experimental study, in which 15 native speakers are asked to rate the acceptability of 1,132 randomly generated NN compounds, with 283 each of the four length patterns. It is found that 2+2 is the most acceptable, followed by 2+1, while 1+2 and 1+1 are the least acceptable. The high rating of 2+2 and the low rating of 1+2 support previous predictions. The rating for 2+1 is slightly lower than expected, and that of 1+1 is most unexpected. It is also found that four other factors (the naturalness of the referent, homograph ambiguity, the boundness of a morpheme, and frequency) have significant effects on acceptability judgment, although their sizes are smaller than that of the length effect. [ABSTRACT FROM AUTHOR]

This paper proposes an average voice model training technique based on a speaker clustering approach to generate synthetic speech with enhanced similarity to the target speakers’ speech. A novel point of the proposed technique is the use of the speaker characteristics (called “speaker class”), which are obtained from unsupervised clustering, as the additional contextual factor for the average voice based speech synthesis. In the model training process, first, speaker clustering is performed for all speakers used for model training to obtain the speaker class for each speaker. The average voice model with multiple speaker characteristics is trained by using the obtained speaker class. For the speaker adaptation and speech parameter generation, the speaker class of the target speaker is estimated on the basis of the Euclidean distance between the centroids of each cluster and the target speaker’s feature. The use of the estimated speaker class makes it possible to utilize the model parameters that have speaker characteristics similar to those of the target speaker for speaker adaptation and speech parameter generation. The results of objective and subjective experiments indicated the proposed technique can synthesize speech with improved similarity and naturalness. [ABSTRACT FROM AUTHOR]

A growing body of linguistic studies is now deploying judgment experiments to probe both syntactic and phonological knowledge. Thus a research question arises as to what kind of judgment format is useful for probing our linguistic knowledge. Against this theoretical background, this study compares two types of phonological judgment experimentation: a scale-based naturalness judgment task and a forced-choice wug test. The current analysis uses the data from two previously published studies on rendaku, a famous voicing phenomenon found in Japanese compound formation, and Lyman's Law, which is known to inhibit rendaku. Although the two tasks at first sight show a close correlation with each other, a detailed examination of the data shows that the forced-choice wug-test reveals the influence of Lyman's Law on rendaku more clearly than the naturalness judgment experiment. To the extent that the effect of Lyman's Law is real, the current comparison shows that a forced-choice wug experiment is better than a naturalness judgment experiment. While the impact of the current results is limited and modest, this study provides a first step toward understanding how different tasks in phonological experimentation may compare to one another. It is hoped that the current study will plant a seed for a research program which addresses which kind of phonological judgment experimentation is best-suited to reveal our phonological knowledge. [ABSTRACT FROM AUTHOR]

In Central Catalan, word-final sibilants and stop + sibilant clusters undergo voicing when followed by a vowel in the next word. In word-final pre-vocalic singleton stops, however, no voicing is observed. This asymmetry is unexpected from the point of view of phonetic naturalness, and it invites a closer empirical investigation, especially in the light of recent phonetic findings which challenge earlier descriptions of a similar voicing process before sonorant consonants in Catalan. This study presents a systematic acoustic comparison of voicing in pre-vocalic stops /p, b/, sibilants /s, z/ and stop + sibilant clusters /bz/ with voicing before obstruents and before sonorant consonants. The results show that while pre-sonorant voicing is limited and highly variable, pre-vocalic sibilant and cluster voicing are both robust categorical processes for most speakers. Drawing on evidence from how voicing is realised in a subset of gradient cases, I propose that pre-vocalic cluster voicing developed from pre-vocalic sibilant voicing which in turn descended from an earlier intervocalic voicing process. I further question whether the outcome of the sound changes in Catalan can be synchronically analysed using grounded markedness constraints, and to what extent the pre-vocalic voicing may be regulated by universal vs. language-specific mechanisms of phonological abstraction. [ABSTRACT FROM AUTHOR]

Highlights: [•] We propose a technique for adding prosodic variations in HMM-based speech synthesis. [•] Phrase-level contexts are defined that compensate the lack of prosodic variations. [•] Average prosodic difference between original and synthetic speech is used. [•] Expressive speech of sales talk and fairy tale is used for evaluation. [•] Users can manually change phrase-level pitch without degrading naturalness. [Copyright &y& Elsevier]

Abstract: This paper analyzes the impacts of machine translation and speech synthesis on speech-to-speech translation systems. A typical speech-to-speech translation system consists of three components: speech recognition, machine translation and speech synthesis. Many techniques have been proposed for integration of speech recognition and machine translation. However, corresponding techniques have not yet been considered for speech synthesis. The focus of the current work is machine translation and speech synthesis, and we present a subjective evaluation designed to analyze their impact on speech-to-speech translation. The results of these analyses show that the naturalness and intelligibility of the synthesized speech are strongly affected by the fluency of the translated sentences. In addition, several features were found to correlate well with the average fluency of the translated sentences and the average naturalness of the synthesized speech. [Copyright &y& Elsevier]

Abstract: Lyman''s Law is a general phonotactic restriction in Japanese which prohibits two voiced obstruents within the same morpheme. This law manifests itself, for example, in the blockage of Rendaku, a phenomenon which voices the initial consonant of the second member of a compound. Lyman''s Law blocks Rendaku when the second member already contains a voiced obstruent. Lyman''s Law has been formulated as a general phonotactic restriction against two voiced obstruents (), and believed to hold only in native words, not in loanwords, because there are many loanwords that violate this restriction (e.g. [gaado] ‘guard’ and [bagu] ‘bug’: ). Building on , and , however, this study shows that Lyman''s Law is active even in loanwords, and nonce words more generally. In Experiments I and II, native speakers of Japanese judged Rendaku in nonce words to be less natural when it resulted in a violation of Lyman''s Law. In Experiment III, native speakers of Japanese judged devoicing of real loanwords and nonce words to be more natural when devoicing was caused by Lyman''s Law. Therefore, the three experiments, as a package, show that Lyman''s Law is active both as a blocker and a trigger of phonological alternations. A general implication of this study is that a restriction with many lexical exceptions can still impact native speakers’ treatment of loanwords and nonce words, as predicted by theories that posit that constraints are violable (). [Copyright &y& Elsevier]

Abstract: The quality of current commercial speech synthesis systems is now so high that system improvements are being made at subtle sub- and supra-segmental levels. Human perceptual evaluation of such subtle improvements requires a highly sophisticated level of perceptual attention to specific acoustic characteristics or cues. However, it is not well understood what acoustic cues listeners attend to by default when asked to evaluate synthetic speech. It may, therefore, be potentially quite difficult to design an evaluation method that allows listeners to concentrate on only one dimension of the signal, while ignoring others that are perceptually more important to them. The aim of the current study was to determine which acoustic characteristics of unit-selection synthetic speech are most salient to listeners when evaluating the naturalness of such speech. This study made use of multidimensional scaling techniques to analyse listeners’ pairwise comparisons of synthetic speech sentences. Results indicate that listeners place a great deal of perceptual importance on the presence of artifacts and discontinuities in the speech, somewhat less importance on aspects of segmental quality, and very little importance on stress/intonation appropriateness. These relative differences in importance will impact on listeners’ ability to attend to these different acoustic characteristics of synthetic speech, and should therefore be taken into account when designing appropriate methods of synthetic speech evaluation. [Copyright &y& Elsevier]

Abstract: For most people “naturalness” is a highly appreciated material characteristic. For instance, a natural wooden floor is seen as more valuable than a fake replica, though they may be comparable in quality and durability. In the present study we investigated how sensory input (vision and touch) contributes to the perception of naturalness in wood. Participants rated samples of wood or imitations thereof, such as vinyl and veneers. We first attempted to provide a validation of the measurement of perceived naturalness by comparing four psychophysical measurement methods (labelled scaling, magnitude estimation, binary decision, and ranked ordering). Second, we investigated the contribution of vision and touch by measuring the perception of naturalness in three exploration modalities (vision only, touch only, and visuo-tactile). The results show a high degree of consistency across measurement methods, suggesting that we measured a common underlying construct that relates to naturalness. It also suggests that this construct is represented on a metathetic (categorical) continuum. Moreover, we found that both vision and touch are highly correlated predictors of visuo-tactile perception of naturalness. [ABSTRACT FROM AUTHOR]

Abstract: The purposes of this study were to investigate naturalness of the post-treatment speech of Comprehensive Stuttering Program (CSP) clients and differences in naturalness ratings by three listener groups. Listeners were 21 student speech-language pathologists, 9 community members, and 15 listeners who stutter. Listeners rated perceptually fluent speech samples of CSP clients obtained immediately post-treatment (Post) and at 5 years follow-up (F5), and speech samples of matched typically fluent (TF) speakers. A 9-point interval rating scale was used. A 3 (listener group)×2 (time)×2 (speaker) mixed ANOVA was used to test for differences among mean ratings. The difference between CSP Post and F5 mean ratings was statistically significant. The F5 mean rating was within the range reported for typically fluent speakers. Student speech-language pathologists were found to be less critical than community members and listeners who stutter in rating naturalness; however, there were no significant differences in ratings made by community members and listeners who stutter. Results indicate that the naturalness of post-treatment speech of CSP clients improves in the post-treatment period and that it is possible for clients to achieve levels of naturalness that appear to be acceptable to adults who stutter and that are within the range of naturalness ratings given to typically fluent speakers. Educational objectives: Readers will be able to (a) summarize key findings of studies that have investigated naturalness ratings, and (b) interpret the naturalness ratings of Comprehensive Stuttering Program speaker samples and the ratings made by the three listener groups in this study. [Copyright &y& Elsevier]

Abstract: The effects of SpeechEasy on stuttering frequency, stuttering severity self-ratings, speech rate, and speech naturalness for 31 adults who stutter were examined. Speech measures were compared for samples obtained with and without the device in place in a dispensing setting. Mean stuttering frequencies were reduced by 79% and 61% for the device compared to the control conditions on reading and monologue tasks, respectively. Mean severity self-ratings decreased by 3.5 points for oral reading and 2.7 for monologue on a 9-point scale. Despite dramatic reductions in stuttering frequency, mean global speech rates in the device condition increased by only 8% in the reading task and 15% for the monologue task, and were well below normal. Further, complete elimination of stuttering was not associated with normalized speech rates. Nevertheless, mean ratings of speech naturalness improved markedly in the device compared to the control condition and, at 3.3 and 3.2 for reading and monologue, respectively, were only slightly outside the normal range. These results show that SpeechEasy produced improved speech outcomes in an assessment setting. However, findings raise the issue of a possible contribution of slowed speech rate to the stuttering reduction effect, especially given participants’ instructions to speak chorally with the delayed signal as part of the active listening instructions of the device protocol. Study of device effects in situations of daily living over the long term is necessary to fully explore its treatment potential, especially with respect to long-term stability. Educational objectives : The reader will be able to discuss and evaluate: (1) issues pertinent to evaluating treatment benefits of fluency aids and (2) the effects of SpeechEasy on stuttering frequency, speech rate, and speech naturalness during testing in a dispensing setting for a relatively large sample of adults who stutter. [Copyright &y& Elsevier]

0It is one of the most basic generalizations in phonology that only certain sets of sounds pattern together in phonological processes. These sets are referred to as natural classes. This paper develops a new analysis of the natural class generalization, formulated in terms of Optimality Theory. It is shown that natural classes derive from the nature of the set of markedness constraints. For example, sounds can pattern together as a natural class if they violate markedness constraints in the same environment, so given constraints *XA and *XB A and B can form a natural class. As a result the range of possible natural classes depends on the inventory of constraints, not on the feature set. This analysis is shown to have empirical advantages over the standard account according to which natural classes are characterized purely in terms of features. [ABSTRACT FROM AUTHOR]