Fernando Llanos

Auditory enhancement and second language experience in Spanish and English weighting of secondary voicing cues

The role of secondary cues in voicing categorization was investigated in three listener groups: Monolingual English (n = 20) and Spanish speakers (n = 20), and Spanish speakers with significant English experience (n = 16). Results showed that, in all three groups, participants used onset f0 in making voicing decisions only in the positive voice onset time (VOT) range (short lag and long lag tokens), while there was no effect of onset f0 on voicing categorization within the negative VOT range (voicing lead tokens) for any of the participant groups. These results support an auditory enhancement view of perceptual cue weighting: Onset f0 serves as a secondary cue to voicing only in the positive VOT range where it is not overshadowed by the presence of pre-voicing. Moreover, results showed that Spanish learners of English gave a significantly greater weight to onset f0 in their voicing decisions than did listeners in either of the other two groups. This result supports the view that learners may overweight secondary cues to distinguish between non-native categories that are assimilated to the same native category on the basis of a primary cue.

Phonological status, not voice onset time, determines the acoustic realization of onset f0 as a secondary voicing cue in Spanish and English

Abstract The covariation of onset f0 with voice onset time (VOT) was examined across and within phonological voicing categories in two languages, English and Spanish. The results showed a significant co-dependency between onset f0 and VOT across phonological voicing categories but not within categories, in both languages. Thus, English short lag and long lag VOT stops, which contrast phonologically, were found to differ significantly in onset f0. Similarly, Spanish short lag and lead VOT tokens are phonologically contrastive and also differed significantly in terms of onset f0. In contrast, English short lag and lead VOT stops, which are sub-phonemic variants of the same phonological category, did not differ in terms of onset f0. These results highlight the importance of phonological factor in determining the pattern of covariation between VOT and onset f0.

The Effects of Language Experience and Speech Context on the Phonetic Accommodation of English-accented Spanish Voicing

Native speakers of Spanish with different amounts of experience with English classified stop-consonant voicing (/b/ versus /p/) across different speech accents: English-accented Spanish, native Spanish, and native English. While listeners with little experience with English classified target voicing with an English- or Spanish-like voice onset time (VOT) boundary, predicted by contextual VOT, listeners familiar with English relied on an English-like VOT boundary in an English-accented Spanish context even in the absence of clear contextual cues to English VOT. This indicates that Spanish listeners accommodated English-accented Spanish voicing differently depending on their degree of familiarization with the English norm.

Hidden Markov modeling of frequency-following responses to Mandarin lexical tones

BACKGROUND The frequency-following response (FFR) is a scalp-recorded electrophysiological potential reflecting phase-locked activity from neural ensembles in the auditory system. The FFR is often used to assess the robustness of subcortical pitch processing. Due to low signal-to-noise ratio at the single-trial level, FFRs are typically averaged across thousands of stimulus repetitions. Prior work using this approach has shown that subcortical encoding of linguistically-relevant pitch patterns is modulated by long-term language experience. NEW METHOD We examine the extent to which a machine learning approach using hidden Markov modeling (HMM) can be utilized to decode Mandarin tone-categories from scalp-record electrophysiolgical activity. We then assess the extent to which the HMM can capture biologically-relevant effects (language experience-driven plasticity). To this end, we recorded FFRs to four Mandarin tones from 14 adult native speakers of Chinese and 14 of native English. We trained a HMM to decode tone categories from the FFRs with varying size of averages. RESULTS AND COMPARISONS WITH EXISTING METHODS Tone categories were decoded with above-chance accuracies using HMM. The HMM derived metric (decoding accuracy) revealed a robust effect of language experience, such that FFRs from native Chinese speakers yielded greater accuracies than native English speakers. Critically, the language experience-driven plasticity was captured with average sizes significantly smaller than those used in the extant literature. CONCLUSIONS Our results demonstrate the feasibility of HMM in assessing the robustness of neural pitch. Machine-learning approaches can complement extant analytical methods that capture auditory function and could reduce the number of trials needed to capture biological phenomena.

Shannon entropy predicts the sonority status of natural classes in English

Speech sounds tend to co-occur in the speech stream according to specific combinatory patterns predicted from their sonority status [Parker, S. G. (2002). Quantifying the sonority hierarchy. Unpublished doctoral dissertation, University of Massachusetts, Amherst, MA]. This study introduces a measure of spectral complexity, inspired by Shannon entropy, that ranks American English phonemes into a minimal version of the sonority hierarchy: vowels > approximants > nasals > fricatives > affricates > stops. Spectral complexity for every consonant and vowel in the TIMIT database was calculated by first parsing the phonemes into 20-ms segments and computing an FFT. For each short-term FFT, Shannon entropy was computed using the distribution of relative amplitudes (dB) across frequency. Average entropy across the FFTs was used to index spectral complexity for the phonemes, which were then sorted by sonority status. Results of a between-group comparison with spectral complexity as the independent variable and natur...

Tracing the Trajectory of Sensory Plasticity across Different Stages of Speech Learning in Adulthood

Although challenging, adults can learn non-native phonetic contrasts with extensive training [1, 2], indicative of perceptual learning beyond an early sensitivity period [3, 4]. Training can alter low-level sensory encoding of newly acquired speech sound patterns [5]; however, the time-course, behavioral relevance, and long-term retention of such sensory plasticity is unclear. Some theories argue that sensory plasticity underlying signal enhancement is immediate and critical to perceptual learning [6, 7]. Others, like the reverse hierarchy theory (RHT), posit a slower time-course for sensory plasticity [8]. RHT proposes that higher-level categorical representations guide immediate, novice learning, while lower-level sensory changes do not emerge until expert stages of learning [9]. We trained 20 English-speaking adults to categorize a non-native phonetic contrast (Mandarin lexical tones) using a criterion-dependent sound-to-category training paradigm. Sensory and perceptual indices were assayed across operationally defined learning phases (novice, experienced, over-trained, and 8-week retention) by measuring the frequency-following response, a neurophonic potential that reflects fidelity of sensory encoding, and the perceptual identification of a tone continuum. Our results demonstrate that while robust changes in sensory encoding and perceptual identification of Mandarin tones emerged with training and were retained, such changes followed different timescales. Sensory changes were evidenced and related to behavioral performance only when participants were over-trained. In contrast, changes in perceptual identification reflecting improvement in categorical percept emerged relatively earlier. Individual differences in perceptual identification, and not sensory encoding, related to faster learning. Our findings support the RHT-sensory plasticity accompanies, rather than drives, expert levels of non-native speech learning.

Power spectral entropy as an information-theoretic correlate of manner of articulation in American English

While all languages differentiate speech sounds by manner of articulation, none of the acoustic correlates proposed to date seem to account for how these contrasts are encoded in the speech signal. The present study describes power spectral entropy (PSE), which quantifies the amount of potential information conveyed in the power spectrum of a given sound. Results of acoustic analyses of speech samples extracted from the Texas Instruments-Massachusetts Institute of Technology database reveal a statistically significant correspondence between PSE and American English major classes of manner of articulation. Thus, PSE accurately captures an acoustic correlate of manner of articulation in American English.

Infants adapt to speaking rate differences in word segmentation

Throughout their development, infants are exposed to varying speaking rates. Thus, it is important to determine whether they are able to adapt to speech at varying rates and recognize target words from continuous speech despite speaking rate differences. To address this question, a series of four experiments were conducted to test whether infants can recognize words in continuous speech when rate is variable. In addition, the underlying mechanisms that infants may use to cope with variations induced by different speaking rates were also examined. Specifically, using the Headturn Preference procedure [Jusczyk and Aslin (1995). Cognitive Psychol. 29, 1-23], infants were familiarized with normal-rate passages containing two trisyllabic target words (e.g., elephants and dinosaurs), and tested with familiar (elephants and dinosaurs) and unfamiliar (crocodiles and platypus) words embedded in normal-rate (experiment 1), fast-rate (experiments 2 and 3), or slow-rate passages (experiment 4). The results indicate that 14-month-olds, but not 11-month-olds, recognized target words in passages with a fast speaking rate. In addition, findings suggest that infants used context to normalize speech across different speaking rates.

Decoding linguistically-relevant pitch patterns from frequency-following responses using hidden Markov models

Pitch encoding is often studied with frequency-following response (FFR), a scalp-recorded potential reflecting phase-locked activity from auditory subcortical ensembles. Prior work using FFR have shown that long-term language experience modulates subcortical encoding of linguistically-relevant pitch patterns. These studies typically rely on FFRs averaging across thousands of repetitions, due to low signal-to-noise ratio of single-trial FFRs. Here, we evaluated the extent to which hidden Markov models (HMMs), with fewer numbers of trials, can be used to quantify pitch encoding as well as capture language experience-dependent plasticity in pitch encoding. FFRs were recorded from fourteen Mandarin Chinese and fourteen American English passively listening to four Mandarin tones (1000 trials per tone). HMMs were used to recognize FFRs to each tone in individual participants. Specifically, HMMs were trained and tested across FFR sets of different sizes, ranging from 50 to 500 trials. Results showed that HMMs we...

Exploiting second-order statistics improves statistical learning of vowels

Unsupervised clustering algorithms were used to evaluate three models of statistical learning of minimal contrasts between English vowel pairs. The first two models employed only first-order statistics with assumptions of uniform [M1] or Gaussian [M2] distributions of vowels in an F1-F2 space. The third model [M3] employed second-order statistics by encoding covariance between F1 and F2. Acoustic measures of F1/F2 frequencies for 12 vowels spoken by 139 men, women, and children (Hillendrand et al. 1995) were used as input to the models. Effectiveness of each model was tested for each minimal-pair contrast across 100 simulations. Each simulation consisted of two centroids that adjusted on a trial-by-trial basis as 1000 F1/F2 pairs were input to the models. With addition of each pair, centroids were reallocated by a k-means algorithm, an unsupervised clustering algorithm that provides an optimal partition of the space into uniformly-sized convex cells. The first-order Gaussian model [M2] performed better th...