Patrick Reidy

Methods for eliciting, annotating, and analyzing databases for child speech development

Methods from automatic speech recognition (ASR), such as segmentation and forced alignment, have facilitated the rapid annotation and analysis of very large adult speech databases and databases of caregiver-infant interaction, enabling advances in speech science that were unimaginable just a few decades ago. This paper centers on two main problems that must be addressed in order to have analogous resources for developing and exploiting databases of young childrens speech. The first problem is to understand and appreciate the differences between adult and child speech that cause ASR models developed for adult speech to fail when applied to child speech. These differences include the fact that childrens vocal tracts are smaller than those of adult males and also changing rapidly in size and shape over the course of development, leading to between-talker variability across age groups that dwarfs the between-talker differences between adult men and women. Moreover, children do not achieve fully adult-like speech motor control until they are young adults, and their vocabularies and phonological proficiency are developing as well, leading to considerably more within-talker variability as well as more between-talker variability. The second problem then is to determine what annotation schemas and analysis techniques can most usefully capture relevant aspects of this variability. Indeed, standard acoustic characterizations applied to child speech reveal that adult-centered annotation schemas fail to capture phenomena such as the emergence of covert contrasts in childrens developing phonological systems, while also revealing childrens nonuniform progression toward community speech norms as they acquire the phonological systems of their native languages. Both problems point to the need for more basic research into the growth and development of the articulatory system (as well as of the lexicon and phonological system) that is oriented explicitly toward the construction of age-appropriate computational models.

A comparison of spectral estimation methods for the analysis of sibilant fricatives

It has been argued that, to ensure accurate spectral feature estimates for sibilants, the spectral estimation method should include a low-variance spectral estimator; however, no empirical evaluation of estimation methods in terms of feature estimates has been given. The spectra of /s/ and /ʃ/ were estimated with different methods that varied the pre-emphasis filter and estimator. These methods were evaluated in terms of effects on two features (centroid and degree of sibilance) and on the detection of four linguistic contrasts within these features. Estimation method affected the spectral features but none of the tested linguistic contrasts.

Moving targets and unsteady states: “Shifting” productions of sibilant fricatives by young children

The English voiceless sibilant /s/--/ʃ/ contrast is one that many children do not acquire until their adolescent years. This protracted acquisition may be due to the high level of articulatory control that is necessary to the successful production of an adult-like sibilant, which involves the coordination of lingual, mandibular, and pulmonic gestures. Poor coordination among these gestures can result in the acoustic properties of the noise source or the vocal tract filter changing throughout the timecourse of the frication, to the extent that the phonetic percept of the frication noise changes across its duration. The present study examined such “shifting” productions of sibilant fricatives by native English-acquiring two- through five-year-old children, which were identified from the Paidologos corpus as those productions where the interval of frication was transcribed phonetically as a sequence of fricative sounds. There were two types of shift in frication quality: (1) a gradual change in the resonant ...

The (null) effect of spectral estimator on the estimation of spectral moments

The spectra of English voiceless sibilants [s] and [s], when computed with traditional estimators, such as the DFT calculated over an interval multiplied with a data window, exhibit relatively large variance, which is believed to introduce error in the estimation of linguistically meaningful features, such as the first four spectral moments (centroid, variance, skewness, and kurtosis). In an effort to reduce this error, it is becoming common practice to compute such features from a multitaper spectrum (MTS)—an estimator, which asymptotically has a fraction of the DFTs variance. However, while the difference in variance has been demonstrated mathematically when the number of data samples approaches infinity, it remains an open question whether the MTS engenders more precise spectral features when estimated from the short intervals that are relevant for comparing [s] and [s]. To evaluate this issue empirically, the first four moments were estimated, with an MTS and a hamming-windowed DFT, from the middle 4...

Quantifying robustness of the /t/-/k/ contrast using a single, static spectral feature

Dynamic spectral shape features accurately classify /t/ and /k/ productions across speakers and contexts. This paper shows that word-initial /t/ and /k/ tokens produced by 21 adults can be differentiated using a single, static spectral feature when spectral energy concentration is considered relative to expectations within a given speaker and vowel context. Centroid and peak frequency-calculated from both acoustic and psychoacoustic spectra-were compared to determine whether one feature could reliably differentiate /t/ and /k/, and, if so, which feature best differentiated them. Centroid frequency from both acoustic and psychoacoustic spectra accurately classified productions of /t/ and /k/.

Learning acoustic features for English stops with graph-based dimensionality reduction

This study applies a semi-supervised graph-based dimensionality reduction algorithm (Laplacian Eigenmaps [Belkin & Nyogi, 2002]) to analyze burst spectra from adult productions of English /k/ and /t/. Multitaper spectra calculated over 25-ms windows were passed through a gammatone filter bank, which models the auditory periphery’s frequency selectivity and frequency-scale compression. From these psychoacoustic spectra, a graph was constructed: node pairs (two spectra) were connected if they shared a common talker or target word, and connecting edges were weighted by the symmetric Kullback-Leibler divergence between the spectra. This graph’s eigenvectors map the spectra into a low-dimensional feature space. Our preliminary experiments with 512 tokens produced by 16 talkers suggest that this algorithm is able to learn a two-dimensional representation of the bursts which reflects well-established articulatory constriction features. The first dimension linearly separated /k/ from /t/ in the back vowel environ...

An electromagnetic articulography study of stop-/s/ and /s/-stop clusters in Greek

Previous kinematic studies on consonant clusters suggest that many factors affect their articulatory timing, including speech rate, frequency of occurrence, and prosody. Other possible factors include the direction of articulatory movement (front-to-back, back-to-front), whether independent articulators (e.g., tongue and lips) or single articulators (e.g., tongue tip and tongue body) are involved, as well as the relative sonority of the consonantal segments. In this study, five talkers of Modern Greek produced the clusters /sp/, /ps/, /sk/, and /ks/ in the carrier phrase “Ipa_______pali” (I said________again). These clusters varied in articulatory direction, articulatory independence, and sonority patterns. We used four methods to determine the degree of articulatory overlap. Each method yielded durations (ms) between gestural landmarks, which were used to compute consonant overlap. The methods differed in the ways gestures were interpreted from the velocity signal (e.g., using extrema or 20% threshold) a...

The differential development of vowel context effects on sibilant fricatives

Previous work has found that vowel-context effects on the static acoustic properties of sibilant fricatives weaken as children age. The current study extended this prior work by analyzing the development of context effects on the spectral dynamics of English sibilants. Native adults and children (2 through 5 years old) produced /s, ʃ/ in a range of pre-vocalic contexts (/i, e, ɑ, o, u/). Effects of vowel rounding and vowel height were investigated through two pyschoacoustic measures computed from auditory spectra: peak ERB number and excitation drop (difference between maximum high-band and minimum low-band excitation). These measures were estimated from 17 20-ms windows spaced evenly across each production. Effects of vowel context on the intercept and shape of the resulting 17-point trajectories were analyzed with polynomial growth-curve models. Context effects were found to differentially weaken or strengthen in children depending on whether it was the intercept or the shape of the trajectory that was affected. For both sibilants, rounding and height effects on the peak-ERB and excitation-drop intercepts generally weakened with age; whereas, height effects on the shape factors of both trajectories tended to strengthen with age. Future work will extend the analysis to Japanese /s, ɕ/.

Steady as /∫i/ goes: The spectral kinematics of sibilant fricatives in English and Japanese

Sibilant fricatives are often treated as having steady-state articulatory targets, which fix their spectra throughout their duration; however, Iskarous et al. (2011) reported that the centroid frequency of English /s/ varies considerably across the fricatives time course. This study replicates their spectral analysis using a psychoacoustic measure (peak ERB) and then extends it to English /∫/ and Japanese /s, ɕ/. The time-varying spectral pattern of each fricative was approximated with a nine-point peak ERB trajectory, computed from 20-ms windows spaced evenly throughout each token. There were three notable results. First, adults did not produce the same spectral kinematic pattern for all sibilants in a given language: the spectral peak of English /s/ followed a concave trajectory, while /∫/ remained relatively flat. Second, phonetically similar fricatives from different languages did not necessarily show similar dynamical spectral patterns: the peak trajectory of /s/ was curved in both languages, but re...