Trained Phoneticians Research Articles

Spectral and temporal implementation of Japanese speakers’ English vowel categories: A corpus-based study

This study investigates the predictions of second language (L2) acquisition models — SLM, PAM-L2, and L2LP — on how native (L1) Japanese speakers implement the spectral and temporal aspects of L2 American English vowel categories. Data were obtained from 102 L1 Japanese speakers in the J-AESOP corpus, which also includes nativelikeness judgments by trained phoneticians. Spectrally, speakers judged to be less nativelike showed a strong influence of L1-L2 category assimilation, except L2 /æː/ whose dissimilarity from L1 categories likely triggered new category formation, in accordance with SLM and PAM-L2. However, more nativelike speakers showed vowel spectra similar to those of native English speakers for all vowels, which aligns better with L2LP. Temporally, although speakers tended to assimilate the phonetic length of English vowels to Japanese phonemic length distinctions, segment-level L1-L2 category similarity was not a significant predictor of the speakers’ nativelikeness. Instead, implementation of prosodic-level factors such as stress and phrase-final lengthening were better predictors. The results highlight the importance of suprasegmental factors in successful L2 category learning and also reveal a weakness in current models of L2 acquisition, which focus primarily on the segmental level. The use of speech corpora for L2 research is discussed.

Weighted entropic associative memory and phonetic learning

The Entropic Associative Memory (EAM) holds declarative but distributed representations of remembered objects. These are characterized as functions from features to discrete values in an abstract amodal space. Memory objects are registered or remembered through a declarative operation; memory recognition is defined as a logical test and cues of objects not contained in the memory are rejected directly without search; and memory retrieval is a constructive operation. In its original formulation, the content of basic memory units or cells was either on or off, hence all stored objects had the same weight or strength. In the present weighted version (W-EAM) we introduce a basic learning mechanism to the effect that the values of the cells used in the representation of an object are reinforced by the memory register operation. As memory cells are shared by different representations, the corresponding associations are reinforced too. The memory system supports a second form of learning: the distributed representation generalizes and renders a large set of potential or latent units that can used for recognizing novel inputs, which can in turn be used for improving the performance of both the deep neural networks used for modelling perception and action, and of the memory operations. This process can be performed recurrently in open-ended fashion and can be used in long term learning. An experiment in the phonetic domain using the Mexican Spanish DIMEx100 Corpus was carried out. This corpus was collected in a controlled noise-free environment, and was transcribed manually by human trained phoneticians, but consists of a relatively small number of utterances. DIMEx100 was used to produced the initial state of the perceptual and motor modules, and for testing the performance of the memory system at such state. Then the incremental learning cycle was modelled using the Spanish CIEMPIESS Corpus, consisting of a very large number of noisy untagged speech utterances collected from radio and TV. The results support the viability of the Weighted Entropic Associative Memory for modelling cognitive processes, such as phonetic representation and learning, for the construction of applications, such as speech recognition and synthesis, and as a computational model of natural memory.

Comparison of auto-contouring and hand-contouring of ultrasound images of the tongue surface

ABSTRACT Contours traced by trained phoneticians have been considered to be the most accurate way to identify the midsagittal tongue surface from ultrasound video frames. In this study, inter-measurer reliability was evaluated using measures that quantified both how closely human-placed contours approximated each other as well as how consistent measurers were in defining the start and end points of contours. High reliability across three measurers was found for all measures, consistent with treating contours placed by trained phoneticians as the ‘gold standard.’ However, due to the labour-intensive nature of hand-placing contours, automatic algorithms that detect the tongue surface are increasingly being used to extract tongue-surface data from ultrasound videos. Contours placed by six automatic algorithms (SLURP, EdgeTrak, EPCS, and three different configurations of the algorithm provided in Articulate Assistant Advanced) were compared to human-placed contours, with the same measures used to evaluate the consistency of the trained phoneticians. We found that contours defined by SLURP, EdgeTrak, and two of the AAA configurations closely matched the hand-placed contours along sections of the image where the algorithms and humans agreed that there was a discernible contour. All of the algorithms were much less reliable than humans in determining the anterior (tongue-tip) edge of tongue contours. Overall, the contours produced by SLURP, EdgeTrak, and AAA should be useable in a variety of clinical applications, subject to spot-checking. Additional practical considerations of these algorithms are also discussed.

Sibilant Fricative Merging in Taiwan Mandarin: An Investigation of Tongue Postures using Ultrasound Imaging.

In Taiwan Mandarin, retroflex [ʂ] is allegedly merging with dental [s], reducing the traditional three-way contrast between sibilant fricatives (i.e., dental [s]-retroflex [ʂ]-alveopalatal [ɕ]) to a two-way contrast. Most of the literature on the observed merging focuses on the acoustic properties and perceptual identification of the sibilants, whereas much less attention has been drawn to the articulatory evidence accounting for the aforementioned sibilant merging. The current study employed ultrasound imaging techniques to uncover the tongue postures for the three sibilant fricatives [s, ʂ, ɕ] in Taiwan Mandarin occurring before vowels [a], [ɨ], and [o]. Results revealed varying classes of the [s-ʂ] merger: complete merging (overlap), no merging (non-overlap), and context-dependent merging (context-dependent overlap, which only occurred before [a]). The observed [s-ʂ] merger was also confirmed by the perceptual identification by trained phoneticians. Center of gravity (CoG), a reliable spectral moment of identifying different sibilant fricatives, was also measured to reflect the articulatory-acoustic correspondence. Results showed that the [s-ʂ] merger varies across speakers and may also be conditioned by vowel contexts and that articulatory mergers may not be entirely reflected in CoG values, suggesting that auxiliary articulatory gestures may be employed to maintain the acoustic contrast.

Pitch accent perception and autistic traits for non-native listeners

This study examines the role of the autistic traits in predicting the prosodic prominence for non-native listeners. The subjects completed the Autism Spectrum Quotient (AQ: Baron-Cohen et al., 2001) questionnaire and participated in the Rapid Prosody Transcription Task (Cole et al., 2010), where they marked the prominent words and prosodic boundaries of Barack Obama’s political speech recordings (475 s in total). The recordings were previously annotated by two trained phoneticians using the ToBI (Tones and Break Indices) conventions for Mainstream American English (Beckman, 1997). Previous studies have shown that native speakers and L2 learners use different acoustic measures as predictors of their perception of prosodic prominence (Pinter et al., 2014; You, 2012). Furthermore, Iao et al. (2017) found that individuals who were less sociable were less able to discriminate foreign lexical pitch difference. Therefore, this study takes intrinsic individual difference into account (assessed by AQ) to see its effect on the perception of pitch accents in English for non-native listeners. The overall results will also speak to issues regarding the development of prosody acquisition and the role of attention in prosodic processing.This study examines the role of the autistic traits in predicting the prosodic prominence for non-native listeners. The subjects completed the Autism Spectrum Quotient (AQ: Baron-Cohen et al., 2001) questionnaire and participated in the Rapid Prosody Transcription Task (Cole et al., 2010), where they marked the prominent words and prosodic boundaries of Barack Obama’s political speech recordings (475 s in total). The recordings were previously annotated by two trained phoneticians using the ToBI (Tones and Break Indices) conventions for Mainstream American English (Beckman, 1997). Previous studies have shown that native speakers and L2 learners use different acoustic measures as predictors of their perception of prosodic prominence (Pinter et al., 2014; You, 2012). Furthermore, Iao et al. (2017) found that individuals who were less sociable were less able to discriminate foreign lexical pitch difference. Therefore, this study takes intrinsic individual difference into account (assessed by AQ) to see its e...

The predictors of foreign-accentedness in the home language of Polish–English bilingual children

We investigated the speech patterns and accentedness of Polish–English bilingual children raised in Great Britain to verify whether their L1 Polish would be perceived as different from that of monolinguals matched for age and socioeconomic status. To this end, Polish-language speech samples of 32 bilinguals and 10 monolinguals (a 3:1 ratio, MAge = 5.79) were phonetically analysed by trained phoneticians and rated by 55 Polish raters, who assessed the degree of native accent, intelligibility, acceptability and perceived age. The results show significant differences in the phonetic performance of bilingual and monolingual children – both in terms of atypical speech patterns uncovered in the phonetic analysis and in terms of the holistic accentedness ratings. We also explored the socio-linguistic predictors of accent ratings in bilingual speech and found that the amount of L1 Polish input was the main predictor of accentedness in children's L1 Polish speech, while L2 English input was marginally significant. (149)

Exploring the acoustic characteristics of individual variation

Studies of the acoustic properties of words often analyze a small subset of words across a large population of speakers. Much of the previous research has not investigated the individual variation produced by a single speaker in large sets of words. The present study analyzes the individual variation produced by a male Western Canadian English speaker, who produced 26,800 English words and 9,600 pseudo-words. All pseudo-words were phonotactically licit and were generated using the software package Wuggy (Keuleers & Brysbaert, 2010). Each word has been force-aligned using the Penn Forced Aligner (Yuan & Liberman, 2008) and then hand corrected by trained phoneticians. We investigate the formant space, word pitch contours, segmental duration, and other acoustic characteristics relevant to classes of segments (such as center of gravity for fricatives). An acoustic comparison is performed between the words and pseudo-words. We explore the acoustic variation of the individual segments produced by this speaker and investigate his individual speech patterns. Finally, we consider the value of delving deeply into productions of a single speaker rather than relying on averaged summaries across a sample a large sample.

A Simplified Vocal Profile Analysis Protocol for the Assessment of Voice Quality and Speaker Similarity

A simplified perceptual protocol for the assessment of voice quality (VQ) is attempted based on the Vocal Profile Analysis (VPA) scheme, with the aim of alleviating typical issues associated with the multidimensionality of VQ and enabling an easy quantification of speaker similarity. Twenty-four non-pathological male speakers (12 monozygotic twin pairs) of Standard Peninsular Spanish were perceptually evaluated by two trained phoneticians using the simplified VPA (SVPA). Based on their perceptual ratings, intra- and inter-rater agreement was measured, and an index of speaker similarity was calculated not only between twin pairs but also between non-twin pairs. For that purpose, one member of each twin pair was compared with a member of a different twin pair. Intra- and inter-rater agreement measures were tested with unweighted and linear weighted kappa. Speaker similarity was measured with simple matching coefficients (SMC). The results show that analysts' internal consistency was very high, whereas inter-rater agreement was found to be strongly setting-dependent. SMCs between speakers indicate that twin pairs are, on average, more similar than non-twin pairs. Agreement results suggest that the proposed SVPA is a reliable protocol for the perceptual characterization of VQ, and SMC results confirm that it can also be a useful tool for the assessment of speaker (dis)similarity. The extraction of a voice quality similarity index shows potential in fields like forensic phonetics, but could also be of interest in related areas of voice research and professional practice.

Speech characteristics in neurofibromatosis type 1

Neurofibromatosis 1 (NF1) is a neurocutaneous-skeletal disorder often accompanied with varying degrees of cognitive and motor problems that potentially affect speech and language. While previous studies have shown that NF1 may be associated with a variety of deviations in the patients' speech, they have not investigated the characteristics in phonetic detail. Our clinical observation that many patients share a distinct voice and manner of speaking led to the primary aim of this study, which was to present a comprehensive description of speech in NF1. A total of 62 patients with NF1 (age range 7-66 years), and a control group of 24 speakers (age range 7-62 years) were evaluated for their speech. The test sessions were recorded and the data were analyzed both by ear and by acoustic measurements. The data were analyzed separately by two trained phoneticians, and a summary was produced after comparing the two independent analyses. Various speech problems were observed among patients with NF1. Individual variation was remarkable, but the deviations were more common and severe in children than in adult patients. In addition, men with NF1 had more speech deviations than women with NF1. Findings include deviations in voice quality, problems in regulating pitch, deviant nasality, misarticulation, and disfluency. We suggest that difficulties in speech, particularly in regulating pitch, may have negative social implications. Our results highlight which components of speech require particular attention in speech therapy for patients with NF1.

Effectiveness of a robust computer assisted pronunciation training tool

A reliable ASR-based pronunciation training tool named STAR (self-training accent reduction) is implemented for native speakers of Korean learning American English. STAR is designed to focus on the most frequent phonemic errors made by Korean adult learners and to provide instantaneous feedback. In order to investigate the effectiveness of STAR, ten Korean participants are recruited for this pilot study. The study consists of three phases: Pre-test, training, and post-test. During the pre-test, the participants read a pre-designed word list containing accent sensitive phonemes into a microphone and participate in one session of training using the STAR system. The participants return for two more sessions of training on the following day. The post-test is administered on the third day, after one additional training session. During the post-test, the participants read the same wordlist as the one administered during the pre-test, and an additional wordlist that they were not trained on. Two trained phoneticians listen and transcribe all recordings to examine whether the participants’ productions are more accurate after the STAR training. The results indicate that most of the participants who practice pronunciation with STAR show improvement in their pronunciation.

Do nasalized fricatives exist?

It has been claimed that nasalized fricatives, even sibilant fricatives, exist in some languages. To assess this claim we sought to discover how much velopharyngeal opening can be present before the acoustic and auditory characteristics of fricatives are affected. Two trained phoneticians produced steady-state voiced and voiceless strong and weak fricatives. The back pressure was intermittently bled with a tube of varying diameter (and thus impedance) inserted into the speaker’s mouth via the buccal sulcus and the gap behind the back molars, simulating nasal leakage. Intraoral pressure (Po) was sampled via a catheter inserted into the pharynx through the nose. The changes in amplitude and quality of frication were analyzed acoustically. We found that (1) a vent area of about 18 mm2 caused auditorily and acoustically noticeable lessening of fricative energy, e.g., sibilants sounded more like nonsibilants, (2) for a given vent aperture, Po was diminished less for voiceless than voiced fricatives (because the voiced fricatives have higher upstream impedance at the glottis) and consequently the same vent aperture impaired the amplitude and quality of frication of voiced fricatives more than voiceless ones. Thus the aerodynamic requirements for fricatives seem to be relatively narrow and unforgiving. [Research supported by Committee on Research, UCB, and by DGICYT, Spain, PB 96-1158.]

Infant vocalizations in response to speech: vocal imitation and developmental change.

Infants' development of speech begins with a language-universal pattern of production that eventually becomes language specific. One mechanism contributing to this change is vocal imitation. The present study was undertaken to examine developmental change in infants' vocalizations in response to adults' vowels at 12, 16, and 20 weeks of age and test for vocal imitation. Two methodological aspects of the experiment are noteworthy: (a) three different vowel stimuli (/a/, /i/, and /u/) were videotaped and presented to infants by machine so that the adult model could not artifactually influence infant utterances, and (b) infants' vocalizations were analyzed both physically, using computerized spectrographic techniques, and perceptually by trained phoneticians who transcribed the utterances. The spectrographic analyses revealed a developmental change in the production of vowels. Infants' vowel categories become more separated in vowel space from 12 to 20 weeks of age. Moreover, vocal imitation was documented, infants listening to a particular vowel produced vocalizations resembling that vowel. A hypothesis is advanced extending Kuhl's native language magnet (NLM) model to encompass infants' speech production. It is hypothesized that infants listening to ambient language store perceptually derived representations of the speech sounds they hear which in turn serve as targets for the production of speech utterances. NLM unifies previous findings on the effects of ambient language experience on infants' speech perception and the findings reported here that short-term laboratory experience with speech is sufficient to influence infants' speech production.

METHODOLOGY IN PHONOLOGICAL INTERLANGUAGE

This paper is a review of current practice in the study of phonological interlanguage as reported in Ioup & Weinberger: Interlanguage Phonology (1987) but supplemented by studies from Argentina and UK. The methodology of some 22 pieces of research is surveyed in terms of subjects, range of phonological features, 'synchronic' and 'diachronic' approaches, techniques and materials, audio-recording and transcription. Assessment of data by trained phoneticians is queried in a discussion on pure vs. applied aims in interlanguage research. If intelligibility of the learners' pronunciation is to be assessed, then untrained judges are more appropriate. The advantages and disadvantages of formal techniques are discussed. Ten stages of a full-scale study of interlanguage phonology are presented.

Intelligibility, phonetic distortion, and listener consensus in the speech of hearing‐impaired talkers

Recorded speech from 29 hearing‐impaired and four normal‐hearing speakers was reviewed by two trained phoneticians. Selected segments of the speech were judged according to 14 phonetic attributes such as consonant manner and place, vowel fronting, and breathiness. Magner intelligibility scores for the hearing‐impaired speakers were also available. A feature‐based perceptual distance measure was constructed and used to specify both the degree of phonetic accuracy of the speech and the level of consensus in listener judgments. Analysis of these data reveals strong direct relationships between phonetic accuracy, Magner intelligibility scores, and listener consensus. Analysis of the consensus data according to feature type indicates that listeners may perceive manner features most reliably, with stress, place, and voicing features perceived with successively less reliability (i.e., consensus). [Work supported by NIH.]

The validity of phonetic transcription: Limitations of a sociolinguistic research tool

ABSTRACTAuditory phonetic transcription is a stock-in-trade of sociolinguists; it is transcriptions, not actual speech, that form the raw data of much of sociolinguistics. Given its importance, it is surprising that phonetic transcription has rarely been examined by sociolinguists from the point of view of its validity and its reliability — despite the existence of a certain amount of discussion in the phonetic literature. Rather, it has been treated as a pretheoretical notion. In this article, we report an experiment that compares the auditory transcriptions of trained phoneticians with physiological data on the same utterances, using the technique ofelectropalatography. The experiment shows that (a) there are intervening factors of a psycho-acoustic nature that impinge on a phonetician's transcription, thus affecting validity; and (b) there is considerable inconsistency, both between phoneticians and between a single phonetician's different attempts at transcribing the same token. Both of these latter factors show that a high degree of reliability cannot be assumed.

Discrimination of duration ratios in Estonian and English

Estonian has been described as having a three‐way quantity distinction among disyllabic words based upon whether the ratio of the durations of syllable 1 to syllable 2 is approximately 2:3, 3:2, or 2:1. The present study was designed to examine the ability of listeners to discriminate among such duration ratios independent of other phonetic factors, such as fundamental frequency or segmental variations. Pairs of noise bursts (with 10 ms rise/fall times) whose durations were in the ratios of 1:2, 2:3, 3:2, and 2:1, were constructed. In half of these noise sequences the combined duration of noise 1 + noise 2 was 350 ms; in the other half the combined duration was 450 ms. Subjects were presented with two such sequences and were required to respond whether the duration ratios in the two sequences were the same or different. Equal numbers of sames and differents were randomly presented to 56 native listeners (and five trained phoneticians). Twenty‐eight listeners were native Estonians (whose data were collected in Tallinn, Estonia) and 28 listeners were native American English speakers (whose data collected in Columbus, OH). Obtained results show that both groups of listeners clearly recognized only two contrastive patterns: 1:2 and 2:3 vs 3:2 and 2:1. In addition, overall sequence duration had a significant effect upon duration ratio discriminations. These results will be discussed in light of possible language group difference, phonetic training differences, and in terms of psychoacoustic limits on the contrastive use of duration in languages.

Performance of an expert spectrogram reader

The subject in our study has spent between 2000–2500 h learning to read speech spectrograms. We tested the subject's ability to identify the phonetic content of broadband speech spectrograms of unknown utterances during eight separate sessions of about 4 h each. The expert was presented with 23 spectrograms of English sentences and sequences of words and nonsense words, and 45 English words embedded in a known carrier phrase. The phonetic labels produced by the expert agreed with the phonetic labels produced by trained phoneticians (who listened to the speech) between 80% and 90% of the time, depending upon the scoring method used. When presented with words in a known carrier phrase, labeling performance improved to about 93%. A linguist who was presented with the phonetic transcriptions produced by the spectrogram reader was unable to identify all of the words in ten of 15 utterances, and missed only a single work in each of the remaining five. A film will be shown of the expert segmenting and labeling two spectrograms of unknown utterances.

Performance of an expert spectrogram reader

The subject in our study has spent between 2000 and 2500 h learning to read speech spectrograms. We tested the subject's ability to identify the phonetic content of broadband speech spectrograms of unknown utterances during eight separate sessions of about 4 h each. The expert was presented with 23 spectrograms of English sentences and sequences of words and nonsense words, and 50 English words embedded in a known carrier phrase. The phonetic labels produced by the expert agreed with the phonetic labels produced by trained phoneticians (who listened to the speech) between 80% and 90% of the time, depending upon the scoring method used. When presented with words in a known carrier phrase, labeling performance improved to about 95%. A linguist who was presented with the phonetic transcriptions produced by the spectrogram reader was able to identify all of the words in 10 of 15 utterances, and missed only a single word in each of the remaining five. A film will be shown of the expert segmenting and labeling two spectrograms of unknown utterances.