Hugo Quené

On multi-level modeling of data from repeated measures designs: a tutorial

Data from repeated measures experiments are usually analyzed with conventional ANOVA. Three well-known problems with ANOVA are the sphericity assumption, the design effect (sampling hierarchy), and the requirement for complete designs and data sets. This tutorial explains and demonstrates multi-level modeling (MLM) as an alternative analysis tool for repeated measures data. MLM allows us to estimate variance and covariance components explicitly. MLM does not require sphericity, it takes the sampling hierarchy into account, and it is capable of analyzing incomplete data. A fictitious data set is analyzed with MLM and ANOVA, and analysis results are compared. Moreover, existing data from a repeated measures design are re-analyzed with MLM, to demonstrate its advantages. Monte Carlo simulations suggest that MLM yields higher power than ANOVA, in particular under realistic circumstances. Although technically complex, MLM is recommended as a useful tool for analyzing repeated measures data from speech research.

Effects of timing regularity and metrical expectancy on spoken-word perception.

Certain types of speech, e.g. lists of words or numbers, are usually spoken with highly regular inter-stress timing. The main hypothesis of this study (derived from the Dynamic Attending Theory) is that listeners attend in particular to speech events at these regular time points. Better timing regularity should improve spoken-word perception. Previous studies have suggested only a weak effect of speech rhythm on spoken-word perception, but the timing of inter-stress intervals was not controlled in these studies. A phoneme monitoring experiment is reported, in which listeners heard lists of disyllabic words in which the timing of the stressed vowels was either regular (with equidistant inter-stress intervals) or irregular. In addition, metrical expectancy was controlled by varying the stress pattern of the target word, as either the same or the opposite of the stress pattern in its preceding words. Resulting reac-tion times show a main effect of timing regularity, but not of metrical expectancy. These results suggest that listeners employ attentional rhythms in spoken-word per-ception, and that regular speech timing improves speech communication.

Multilevel modeling of between-speaker and within-speaker variation in spontaneous speech tempo

Speech tempo (articulation rate) varies both between and within speakers. The present study investigates several factors affecting tempo in a corpus of spoken Dutch, consisting of interviews with 160 high-school teachers. Speech tempo was observed for each phrase separately, and analyzed by means of multilevel modeling of the speakers sex, age, country, and dialect region (between speakers) and length, sequential position of phrase, and autocorrelated tempo (within speakers). Results show that speech tempo in this corpus depends mainly on phrase length, due to anticipatory shortening, and on the speakers country, with different speaking styles in The Netherlands (faster, less varied) and in Flanders (slower, more varied). Additional analyses showed that phrase length itself is shorter in The Netherlands than in Flanders, and decreases with speakers age. Older speakers tend to vary their phrase length more (within speakers), perhaps due to their accumulated verbal proficiency.

What Makes Speech Sound Fluent? The Contributions of Pauses, Speed and Repairs.

The oral fluency level of an L2 speaker is often used as a measure in assessing language proficiency. The present study reports on four experiments investigating the contributions of three fluency aspects (pauses, speed and repairs) to perceived fluency. In Experiment 1 untrained raters evaluated the oral fluency of L2 Dutch speakers. Using specific acoustic measures of pause, speed and repair phenomena, linear regression analyses revealed that pause and speed measures best predicted the subjective fluency ratings, and that repair measures contributed only very little. A second research question sought to account for these results by investigating perceptual sensitivity to acoustic pause, speed and repair phenomena, possibly accounting for the results from Experiment 1. In Experiments 2–4 three new groups of untrained raters rated the same L2 speech materials from Experiment 1 on the use of pauses, speed and repairs. A comparison of the results from perceptual sensitivity (Experiments 2–4) with fluency perception (Experiment 1) showed that perceptual sensitivity alone could not account for the contributions of the three aspects to perceived fluency. We conclude that listeners weigh the importance of the perceived aspects of fluency to come to an overall judgment.

On the just noticeable difference for tempo in speech

Abstract Speakers vary their speech tempo (speaking rate), and such variations in tempo are quite noticeable. But what is the just noticeable difference (JND) for tempo in speech? The present study aims at providing a realistic and robust estimate, by using multiple speech tokens from multiple speakers. The JND is assessed in two (2IAX and 2IFC) comparison experiments, yielding an estimated JND for speech tempo of about 5%. A control experiment suggests that this finding is not due to acoustic artefacts of the tempo-transformation method used. Tempo variations within speakers typically exceed this JND, which makes such variations relevant in speech communication.

Word-level intelligibility of time-compressed speech: prosodic and segmental factors

In this study we investigate whether speakers, in line with the predictions of the Hyper- and Hypospeech theory, speed up most during the least informative parts and less during the more informative parts, when they are asked to speak faster. We expected listeners to benefit from these changes in timing, and our main goal was to find out whether making the temporal organisation of artificially time-compressed speech more like that of natural fast speech would improve intelligibility over linear time compression. Our production study showed that speakers reduce unstressed syllables more than stressed syllables, thereby making the prosodic pattern more pronounced. We extrapolated fast speech timing to even faster rates because we expected that the more salient prosodic pattern could be exploited in difficult listening situations. However, at very fast speech rates, applying fast speech timing worsens intelligibility. We argue that the non-uniform way of speeding up may not be due to an underlying communicative principle, but may result from speakers inability to speed up otherwise. As both prosodic and segmental information contribute to word recognition, we conclude that extrapolating fast speech timing to extremely fast rates distorts this balance between prosodic and segmental information. 2002 Elsevier B.V. All rights reserved.

Native speakers' perceptions of fluency and accent in L2 speech

Oral fluency and foreign accent distinguish L2 from L1 speech production. In language testing practices, both fluency and accent are usually assessed by raters. This study investigates what exactly native raters of fluency and accent take into account when judging L2. Our aim is to explore the relationship between objectively measured temporal, segmental and suprasegmental properties of speech on the one hand, and fluency and accent as rated by native raters on the other hand. For 90 speech fragments from Turkish and English L2 learners of Dutch, several acoustic measures of fluency and accent were calculated. In Experiment 1, 20 native speakers of Dutch rated the L2 Dutch samples on fluency. In Experiment 2, 20 different untrained native speakers of Dutch judged the L2 Dutch samples on accentedness. Regression analyses revealed, first, that acoustic measures of fluency were good predictors of fluency ratings. Second, segmental and suprasegmental measures of accent could predict some variance of accent ratings. Third, perceived fluency and perceived accent were only weakly related. In conclusion, this study shows that fluency and perceived foreign accent can be judged as separate constructs.

Coping with gradient forms of /t/-deletion and lexical ambiguity in spoken word recognition

This study investigates how listeners cope with gradient forms of deletion of word-final /t/ when recognising words in a phonological context that makes /t/-deletion viable. A corpus study confirmed a high incidence of /t/-deletion in an /st#b/ context in Dutch. A discrimination study showed that differences between released /t/, unreleased /t/ and fully deleted /t/ in this specific /st#b/ context were salient. Two on-line experiments were carried out to investigate whether lexical activation might be affected by this form variation. Even though unreleased and released variants were processed equally fast by listeners, a detailed analysis of the unreleased condition provided evidence for gradient activation. Activating a target ending in /t/ is slowest for the most reduced variant because phonological context has to be taken into account. Importantly, activation for a target with /t/ in the absence of cues for /t/ is reduced if there is a surface-matching lexical competitor.

Acoustic correlates of vowel intelligibility in clear and conversational speech for young normal-hearing and elderly hearing-impaired listeners

The present investigation carried out acoustic analyses of vowels in clear and conversational speech produced by 41 talkers. Mixed-effects models were then deployed to examine relationships among acoustic and perceptual data for these vowels. Acoustic data include vowel duration, steady-state formant frequencies, and two measures of dynamic formant movement. Perceptual data consist of vowel intelligibility in noise for young normal-hearing and elderly hearing-impaired listeners, as reported by Ferguson in 2004 and 2012 [J. Acoust. Soc. Am. 116, 2365-2373 (2004); J. Speech Lang. Hear. Res. 55, 779-790 (2012)], respectively. Significant clear speech effects were observed for all acoustic metrics, although not all measures changed for all vowels and considerable talker variability was observed. Mixed-effects analyses revealed that the contribution of duration and steady-state formant information to vowel intelligibility differed for the two listener groups. This outcome is consistent with earlier research suggesting that hearing loss, and possibly aging, alters the way acoustic cues are used for identifying vowels.

Non-native durational patterns decrease speech intelligibility

In native speech, durational patterns convey linguistically relevant phenomena such as phrase structure, lexical stress, rhythm, and word boundaries. The lower intelligibility of non-native speech may be partly due to its deviant durational patterns. The present study aims to quantify the relative contributions of non-native durational patterns and of non-native speech sounds to intelligibility. In a Speech Reception Threshold study, duration patterns were transplanted between native and non-native versions of Dutch sentences. Intelligibility thresholds (critical speech-to-noise ratios) differed by about 4dB between the matching versions with unchanged durational patterns. Results for manipulated versions suggest that about 0.4-1.1dB of this difference was due to the durational patterns, and that this contribution was larger if the native and non-native patterns were more deviant. The remainder of the difference must have been due to non-native speech sounds in these materials. This finding supports recommendations to attend to durational patterns as well as native-like speech sounds, when learning to speak a foreign language.