Noah H. Silbert

Decisional separability, model identification, and statistical inference in the general recognition theory framework.

Recent work in the general recognition theory (GRT) framework indicates that there are serious problems with some of the inferential machinery designed to detect perceptual and decisional interactions in multidimensional identification and categorization (Mack, Richler, Gauthier, & Palmeri, 2011). These problems are more extensive than previously recognized, as we show through new analytic and simulation-based results indicating that failure of decisional separability is not identifiable in the Gaussian GRT model with either of two common response selection models. We also describe previously unnoticed formal implicational relationships between seemingly distinct tests of perceptual and decisional interactions. Augmenting these formal results with further simulations, we show that tests based on marginal signal detection parameters produce unacceptably high rates of incorrect statistical significance. We conclude by discussing the scope of the implications of these results, and we offer a brief sketch of a new set of recommendations for testing relationships between dimensions in perception and response selection in the full-factorial identification paradigm.

Focus, prosodic context, and phonological feature specification : Patterns of variation in fricative production

Because they consist, in large part, of random turbulent noise, fricatives present a challenge to attempts to specify the phonetic correlates of phonological features. Previous research has focused on temporal properties, acoustic power, and a variety of spectral properties of fricatives in a number of contexts [Jongman et al., J. Acoust. Soc. Am. 108, 1252-1263 (2000); Jesus and Shadle, J. Phonet. 30, 437-467 (2002); Crystal and House, J. Acoust. Soc. Am. 83, 1553-1573 (1988a)]. However, no systematic investigation of the effects of focus and prosodic context on fricative production has been carried out. Manipulation of explicit focus can serve to selectively exaggerate linguistically relevant properties of speech in much the same manner as stress [de Jong, J. Acoust. Soc. Am. 97, 491-504 (1995); de Jong, J. Phonet. 32, 493-516 (2004); de Jong and Zawaydeh, J. Phonet. 30, 53-75 (2002)]. This experimental technique was exploited to investigate acoustic power along with temporal and spectral characteristics of American English fricatives in two prosodic contexts, to probe whether native speakers selectively attend to subsegmental features, and to consider variability in fricative production across speakers. While focus in general increased noise power and duration, speakers did not selectively enhance spectral features of the target fricatives.

Independence and separability in the perception of complex nonspeech sounds

All sounds are multidimensional, yet the relationships among auditory dimensions have been studied only infrequently. General recognition theory (GRT; Ashby & Townsend, 1986) is a multidimensional generalization of signal detection theory and, as such, provides powerful tools well suited to the study of the relationships among perceptual dimensions. However, previous uses of GRT have been limited in serious ways. We present methods designed to overcome these limitations, and we use these methods to apply GRT to investigations of the relationships among auditory perceptual dimensions that previous work suggests are independent (frequency, duration) or not (fundamental frequency [ f0], spectral shape). Results from Experiment 1 confirm that frequency and duration do not interact decisionally, and they extend this finding with evidence of perceptual independence. Results from Experiment 2 show that f0 and spectral shape tend to interact perceptually, decisionally, or both, and that perceptual interactions occur within, but not between, stimuli (i.e., the interactions suggest correlated noise across processing channels corresponding to perceptually separable dimensions). The results are discussed in relation to lower level sensory modeling and higher level cognitive and linguistic issues.

Syllable structure and integration of voicing and manner of articulation information in labial consonant identification

Speech perception requires the integration of information from multiple phonetic and phonological dimensions. A sizable literature exists on the relationships between multiple phonetic dimensions and single phonological dimensions (e.g., spectral and temporal cues to stop consonant voicing). A much smaller body of work addresses relationships between phonological dimensions, and much of this has focused on sequences of phones. However, strong assumptions about the relevant set of acoustic cues and/or the (in)dependence between dimensions limit previous findings in important ways. Recent methodological developments in the general recognition theory framework enable tests of a number of these assumptions and provide a more complete model of distinct perceptual and decisional processes in speech sound identification. A hierarchical Bayesian Gaussian general recognition theory model was fit to data from two experiments investigating identification of English labial stop and fricative consonants in onset (syllable initial) and coda (syllable final) position. The results underscore the importance of distinguishing between conceptually distinct processing levels and indicate that, for individual subjects and at the group level, integration of phonological information is partially independent with respect to perception and that patterns of independence and interaction vary with syllable position.

Automated Vocal Analysis of Children With Hearing Loss and Their Typical and Atypical Peers.

Objectives: This study investigated automatic assessment of vocal development in children with hearing loss compared with children who are typically developing, have language delays, and have autism spectrum disorder. Statistical models are examined for performance in a classification model and to predict age within the four groups of children. Design: The vocal analysis system analyzed 1913 whole-day, naturalistic acoustic recordings from 273 toddlers and preschoolers comprising children who were typically developing, hard of hearing, language delayed, or autistic. Results: Samples from children who were hard of hearing patterned more similarly to those of typically developing children than to the language delayed or autistic samples. The statistical models were able to classify children from the four groups examined and estimate developmental age based on automated vocal analysis. Conclusions: This work shows a broad similarity between children with hearing loss and typically developing children, although children with hearing loss show some delay in their production of speech. Automatic acoustic analysis can now be used to quantitatively compare vocal development in children with and without speech-related disorders. The work may serve to better distinguish among various developmental disorders and ultimately contribute to improved intervention.

Fidelity of Automatic Speech Processing for Adult and Child Talker Classifications.

Automatic speech processing (ASP) has recently been applied to very large datasets of naturalistically collected, daylong recordings of child speech via an audio recorder worn by young children. The system developed by the LENA Research Foundation analyzes childrens speech for research and clinical purposes, with special focus on of identifying and tagging family speech dynamics and the at-home acoustic environment from the auditory perspective of the child. A primary issue for researchers, clinicians, and families using the Language ENvironment Analysis (LENA) system is to what degree the segment labels are valid. This classification study evaluates the performance of the computer ASP output against 23 trained human judges who made about 53,000 judgements of classification of segments tagged by the LENA ASP. Results indicate performance consistent with modern ASP such as those using HMM methods, with acoustic characteristics of fundamental frequency and segment duration most important for both human and machine classifications. Results are likely to be important for interpreting and improving ASP output.

Precision and error of automatic speech recognition

1aSCb1. Precision and error of automatic speech recognition. Mark VanDam* and Noah H. Silbert ​ ​ *Corresponding authors address: Speech and Hearing Sciences, Washington State University, Health Sciences Bldg, Room 125S, Spokane, WA 99202, mark.vandam@wsu.edu Experienced judges assessed performance of an automatic speech recognition system developed for linguistic exchanges within families in their natural environment. Preliminary results suggest overall good performance with relatively high precision and low error.

Technical clarification to Silbert and Thomas (2013): “Decisional separability, model identification, and statistical inference in the general recognition theory framework”

We offer a minor technical correction to the published proof of part (ii) of the main theorem in Silbert and Thomas (Psychonomic Bulletin & Review, 20, 1–20, 2013) that somewhat limits the scope of the equivalence observed in that work. Specifically, in order for a mean shift integrality with decisional separability to be mimicked by a perceptually separable but nondecisionally separable configuration, one needs to assume stimulus invariance. This holds when all of the covariance matrices in the stimulus configuration are equal to each other. We note that part (i) of the theorem is unaffected by this modification; an empirical finding of perceptual separability and the failure of decisional separability can be mimicked by a perceptually nonseparable, decisionally separable configuration without restricting the covariance matrices to be equal. We also note that stimulus invariance is often assumed in simple designs (e.g., Macmillan & Ornstein in Journal of the Acoustical Society of America, 97, 1261–1285, 1998), due to the implausibility of different perceptual correlations being present within stimuli perched very closely in perceptual space.

Non-native phonemic discrimination, phonological short term memory, and word learning

Abstract Accurately perceiving non-native speech sounds is known to be very difficult. Numerous studies provide strong and converging evidence that this difficulty varies systematically, depending on the properties of the non-native sounds and the native language of the listener. There is substantially less research on how phonetic and phonological structure relates to individual differences in the ability to perceive non-native phonemes, though individual differences in auditory abilities are well-documented. The present work reports two experiments aimed at elucidating the structure of individual differences in non-native speech perception and the relationship between these abilities, phonological short term memory, and early second language word learning. In Experiment 1, confirmatory factor analysis of discrimination data for nine non-native contrasts from different languages indicates that voicing and place (segmental) contrasts pattern together and distinct from tone (suprasegmental) contrasts. In Experiment 2, the results indicate that phonological short term memory and discrimination ability both predict word learning accuracy and that discrimination ability does so in a mostly feature-specific manner.

Integration of phonological information in obstruent consonant identification.

Speech perception requires the integration of information from multiple phonetic and phonological dimensions. Numerous studies have investigated the mapping between multiple acoustic‐phonetic dimensions and single phonological dimensions (e.g., spectral and temporal properties of stop consonants in voicing contrasts). Many fewer studies have addressed relationships between phonological dimensions. Most such studies have focused on the perception of sequences of phones (e.g., bid, bed, bit, and bet), though some have focused on multiple phonological dimensions within phones (e.g., voicing and place of articulation in [p], [b], [t], and [d]). However, strong assumptions about relevant acoustic‐phonetic dimensions and/or the nature of perceptual and decisional information integration limit previous findings in important ways. New methodological developments in the general recognition theory framework enable a number of these assumptions to be tested and provide a more complete model of distinct perceptual an...