Automated vocal analysis of naturalistic recordings from children with autism, language delay, and typical development.

For generations the study of vocal development and its role in language has been conducted laboriously, with human transcribers and analysts coding and taking measurements from small recorded samples. Our research illustrates a method to obtain measures of early speech development through automated analysis of massive quantities of day-long audio recordings collected naturalistically in children's homes. A primary goal is to provide insights into the development of infant control over infrastructural characteristics of speech through large-scale statistical analysis of strategically selected acoustic parameters. In pursuit of this goal we have discovered that the first automated approach we implemented is not only able to track children's development on acoustic parameters known to play key roles in speech, but also is able to differentiate vocalizations from typically developing children and children with autism or language delay. The method is totally automated, with no human intervention, allowing efficient sampling and analysis at unprecedented scales. The work shows the potential to fundamentally enhance research in vocal development and to add a fully objective measure to the battery used to detect speech-related disorders in early childhood. Thus, automated analysis should soon be able to contribute to screening and diagnosis procedures for early disorders, and more generally, the findings suggest fundamental methods for the study of language in natural environments.

Beyond ba-ba and gu-gu: challenges and strategies in coding infant vocalizations.

Infant vocal behaviors are extremely complex. Consequently, coding these behaviors is difficult and is typically associated with low reliability across observers. Various difficulties that arise when dealing with prelinguistic vocalizations, especially in the first 6 months of life, are outlined here. A proposed database of digitized infant vocalizations that illustrates strategies used to deal with these difficulties is then described. These strategies are based on theoretical infraphonological constructs, empirical observations, and information about the nature of mature phonological systems. Furthermore, the strategies are open-ended and can be modified as new information becomes available regarding infant vocal behaviors. At present, a preliminary database is available on the Web that illustrates some of these strategies. As the database is expanded, it is expected to provide a general framework for observers to categorize infant vocalizations and thereby enhance observer reliability.

Vocal atypicalities of preverbal autistic children.

This study was designed to evaluate the nature of early vocal behaviors in young children with autism. Recent methodological and conceptual advances in the study of infant preverbal vocalizations were used to provide a detailed examination of the vocal behavior of young preverbal children with autism and comparison children with developmental delays. Results revealed that children with autism did not have difficulty with the expression of well-formed syllables (i.e., canonical babbling). However, children with autism did display significant impairments in vocal quality (i.e., atypical phonation). Specifically, autistic children produced a greater proportion of syllables with atypical phonation than did comparison children. Consistent with prior reports, the children with autism also displayed a deficit in joint attention behaviors. Furthermore, the atypicalities in the vocal behavior of children with autism appeared to be independent of individual differences in joint attention skill, suggesting that a multiple process model may be needed to describe early social-communication impairments in children with autism. Data are discussed in terms of their implications for future theoretical and applied research, including efforts to enhance the specificity of early diagnostic procedures.

Precursors to speech in infancy: the prediction of speech and language disorders.

During the canonical stage of infant babbling, infants produce well-formed syllables, often in reduplicated sequences such as "bababa." Although nearly all infants with normal hearing begin the canonical stage by 10 months of age, a few are delayed, and these infants may be of special interest. Recent studies indicate that late onset of canonical babbling may be a predictor of disorders. A simple screening procedure that focuses on canonical babbling was used to evaluate over 3400 infants at risk who were about 10 months of age. Among infants who showed late onset of canonical babbling, fewer than half had been previously diagnosed as having a significant medical problem that might have accounted for the delay. A follow-up study indicated that infants with delayed canonical babbling had smaller production vocabularies at 18, 24, and 30 months than did infants in the control group. The results suggest that late onset of canonical babbling, a factor that can be monitored effectively through an interview with a parent, can predict delay in the onset of speech production.

Use of technology in phonological assessment: evaluation of early meaningful speech and prelinguistic vocalizations.

Technologies that assist in the assessment of phonological abilities in early meaningful speech and prelinguistic vocalizations are discussed. Computerized assessment of phonological abilities (CAP) software offers significant time savings for basic analyses and makes possible the completion of more detailed analyses. The programs discussed in this article allow flexibility in determining not only the most appropriate analyses for an individual client, but also the design of the optimal symbol system for representing the client's productions. Case studies are included to illustrate the clinical benefits of CAP technology.

An event-based analysis of the coordination of early infant vocalizations and facial actions.

This study used an event-based approach to provide empirical evidence regarding the nature of coordination in 3- and 6-month-old infants. Vocalizations and facial actions of 12 normally developing infants interacting with their caregivers were coded. Coded vocalizations and facial actions were considered coordinated when they temporally overlapped. Results indicate that infants coordinated their vocalizations and facial actions more than expected by chance. Coordinated events were governed by 2 sequence patterns. When 2 communicative events were temporally associated across modalities, 1 event tended to be completely embedded within the other, and vocalizations tended to end before facial actions. This study provides new information about how infant communication is structured, confirms results from other coordination studies, and describes a new method for analysis of event-based data.

Late onset canonical babbling: a possible early marker of abnormal development.

By their 10th month of life, typically developing infants produce canonical babbling, which includes the well-formed syllables required for meaningful speech. Research suggests that emerging speech or language-related disorders might be associated with late onset of canonical babbling. Onset of canonical babbling was investigated for 1,536 high-risk infants, at about 10-months corrected age. Parental report by open-ended questionnaire was found to be an efficient method for ascertaining babbling status. Although delays were infrequent, they were often associated with genetic, neurological, anatomical, and/or physiological abnormalities. Over half the cases of late canonical babbling were not, at the time they were discovered associated with prior significant medical diagnoses. Late canonical-babbling onset may be a predictor of later developmental disabilities, including problems in speech, language, and reading.

Development of precursors to speech in infants exposed to two languages.

The study of bilingualism has often focused on two contradictory possibilities: that the learning of two languages may produce deficits of performance in each language by comparison with performance of monolingual individuals, or on the contrary, that the learning of two languages may produce linguistic or cognitive advantages with regard to the monolingual learning experience. The work reported here addressed the possibility that the very early bilingual experience of infancy may affect the unfolding of vocal precursors to speech. The results of longitudinal research with 73 infants aged 0;4 to 1;6 in monolingual and bilingual environments provided no support for either a bilingual deficit hypothesis nor for its opposite, a bilingual advantage hypothesis. Infants reared in bilingual and monolingual environments manifested similar ages of onset for canonical babbling (production of well-formed syllables), an event known to be fundamentally related to speech development. Further, quantitative measures of vocal performance (proportion of usage of well-formed syllables and vowel-like sounds) showed additional similarities between monolingual and bilingual infants. The similarities applied to infants of middle and low socio-economic status and to infants that were born at term or prematurely. The results suggest that vocal development in the first year of life is robust with respect to conditions of rearing. The biological foundations of speech appear to be such as to resist modifications in the natural schedule of vocal development.

Longitudinal speech perception performance of young children with cochlear implants and tactile aids plus hearing aids.

Nineteen profoundly deaf children who utilize either cochlear implants (CIs) or multichannel tactile aids plus hearing aids (T + HAs) and who were enrolled in a full-day educational program that specializes in the use of sensory aids were evaluated using a battery of speech perception tests either developed in house or chosen because they were part of a standard battery of tests developed for children with sensory aids. The tests were organized into four perceptual levels ranging from pattern perception at level one to open set word identification at level four. For each level, data were analyzed for changes over time and for differences between performance of CI children and those using T + HAs. The results indicate that overall, across levels, both groups improved significantly over time but no differences were found between users of T + HAs and CIs at any level.

Practical evaluation procedure to assess and remediate speech perception skills.

The University of Miami/Dade County Public Schools Model Program for the Deaf and Hard of Hearing is a research and training effort dedicated to the utilization of sensory aids including hearing aids, tactual vocoders, and cochlear implants. The program's teachers and clinicians follow the Miami Cochlear Implant, Auditory, and Tactile Skills (CHATS) Curriculum for the development of individualized speech perception and production goals. A series of peech perception tests has been used for the past five years to evaluate the children's progress. The test battery, administered at six month intervals, is extensive and impractical for school clinicians and teachers to administer to their students. To assist teachers and clinicians in the process of selecting appropriate goals and objectives for sensory aid training, a speech perception test has been developed to accompany the curriculum. This paper includes a discussion of the test design as it correlates with the curriculum.

A longitudinal evaluation of the speech perception capabilities of children using multichannel tactile vocoders.

Thirty children (mean age 6:11, range 4:3 to 11:0, SD = 2:3) with profound hearing impairments were followed longitudinally over a 3-year period and evaluated every 6 months with a battery of speech perception tests. The battery spanned several levels of perception, from pattern perception to open-set word recognition. The children were all enrolled in a single full-day educational program that used multichannel tactile aids in addition to hearing aids. Testing was conducted in Auditory alone (A), Tactile plus Auditory (TA), Tactile alone (T), and in one instance, Tactile plus Auditory plus Vision (TAV) conditions because the primary interest of the work was the relationship between auditory and tactile training on perception. Results indicated that children's performance improved with age, with the oldest children achieving open-set speech recognition in the TA condition. Performance in the TA condition generally exceeded that in both A and T conditions. Outcomes were compared to those from two studies in the literature for children of similar age with cochlear implants and tactile aids on the same tests. Results suggest that performance of children who had cochlear implants for an average of 21 months was similar to TA and TAV performance of children in the present study who had tactile experience over a similar period.

Relations of motor and vocal milestones in typically developing infants and infants with Down syndrome.

We measured the ages at which typically developing infants and infants with Down syndrome achieved an important vocal milestone (canonical babbling), a rhythmic motor milestone (hand-banging), and six other motor milestones. The interrelations of the milestone onsets, and their relations with Down syndrome, were assessed quantitatively. Hand-banging and canonical babbling were associated and were somewhat delayed by Down syndrome. Stepping, standing, sitting, and creeping/crawling were associated and were severely delayed by Down syndrome. Rolling and reaching were also delayed by Down syndrome, though they were not strongly associated with other milestones or with one another. These results suggest that the rhythmic behaviors (canonical babbling and hand-banging) may be internally linked by common neuromuscular underpinnings and that the postural behaviors may be similarly linked.

Onset of speech-like vocalizations in infants with Down syndrome.

Canonical babbling of infants with and without Down syndrome was compared. Infants with Down syndrome and typically developing infants began canonical babbling in the first year of life, but the infants with Down syndrome began 2 months later. Once begun, their canonical babbling was less stable than that of typically developing infants. Age at onset of canonical babbling for the infants with Down syndrome was correlated with their scores at 27 months of age on the Early Social-Communication Scales. The results of this study suggest that Down syndrome influences vocal development in the first year of life and that early vocal development is related, possibly in combination with motoric and cognitive factors, to later social and communicative functioning of children with Down syndrome.

Cross-language synonyms in the lexicons of bilingual infants: one language or two?

This study tests the widely-cited claim from Volterra & Taeschner (1978), which is reinforced by Clark's PRINCIPLE OF CONTRAST (1987), that young simultaneous bilingual children reject cross-language synonyms in their earliest lexicons. The rejection of translation equivalents is taken by Volterra & Taeschner as support for the idea that the bilingual child possesses a single-language system which includes elements from both languages. We examine first the accuracy of the empirical claim and then its adequacy as support for the argument that bilingual children do not have independent lexical systems in each language. The vocabularies of 27 developing bilinguals were recorded at varying intervals between ages 0;8 and 2;6 using the MacArthur CDI, a standardized parent report form in English and Spanish. The two single-language vocabularies of each bilingual child were compared to determine how many pairs of translation equivalents (TEs) were reported for each child at different stages of development. TEs were observed for all children but one, with an average of 30% of all words coded in the two languages, both at early stages (in vocabularies of 2-12 words) and later (up to 500 words). Thus, Volterra & Taeschner's empirical claim was not upheld. Further, the number of TEs in the bilinguals' two lexicons was shown to be similar to the number of lexical items which co-occurred in the monolingual lexicons of two different children, as observed in 34 random pairings for between-child comparisons. It remains to be shown, therefore, that the bilinguals' lexicons are not composed of two independent systems at a very early age. Furthermore, the results appear to rule out the operation of a strong principle of contrast across languages in early bilingualism.

Phrasing in prelinguistic vocalizations.

Phrasing is a universal characteristic of human communication, and the present investigation explored its developmental roots in nonvegetative, prelinguistic vocalizations. Adult judges identified a hierarchical arrangement of syllables embedded within utterances and utterances embedded within prelinguistic phrases in the vocalizations of infants. Prelinguistic phrases were characterized by systematic lengthening of phrase final syllables, temporal patterning, and stable durations across development that were similar to those of some cross-culturally optimal rhythmic units from other domains. Analyses of vocalizations of infants with Down syndrome indicated similar internal structure of prelinguistic phrases to those of typically developing infants, but with longer durations. These findings suggest that relative durational characteristics of prelinguistic phrasing are stable features of early vocal behavior, although the absolute durational characteristics of prelinguistic phrases can be impacted by a complex and severe disorder such as Down syndrome.

Early vocal development in tactually aided children with severe-profound hearing loss.

The vocalizations of six children with severe-profound hearing loss were audio-recorded in two conditions during individual speech-language intervention sessions: (a) auditory amplification alone, and (b) auditory amplification plus the Tactaid II, a two-channel vibrotactile device (Franklin, 1986). Utterances were categorized according to the infraphonological framework described by Oller (1980, 1986) and Oller and Lynch (1992). Vocalizations were categorized in a developmental framework relative to mature speech. Those utterances containing well-formed consonant-vowel syllables were transcribed with broad phonetic transcription and analyzed at both the syllabic and segmental levels. Statistically significant differences were found between the two conditions for vocal volubility (i.e., quantity of vocalizations produced); subjects vocalized more when using both auditory amplification and the Tactaid II together than with auditory amplification alone. Trends in the early vocal development of these children with severe-profound hearing loss are described at the infraphonologic, segmental, and syllabic levels.

Speech-like vocalizations in infancy: an evaluation of potential risk factors.

This work reports longitudinal evaluation of the speech-like vocal development of infants born at risk due to prematurity or low socio-economic status (SES) and infants not subject to such risk. Twenty infants were preterm (10 of low SES) and 33 were full term (16 of low SES), and all were studied from 0;4 through 1;6. The study provides the indication that at-risk infants are not generally delayed in the ability to produce well-formed speech-like sounds as indicated in tape-recorded vocal samples. At the same time, premature infants show a tendency to produce well-formed syllables less consistently than full terms after the point at which parents and laboratory personnel note the onset of the canonical babbling stage (the point after which well-formed syllables are well established in the infant vocal repertoires). Further, even though low SES infants produce well-formed speech-like structures on schedule, they show a reliably lower tendency to vocalize in general, as reflected by fewer utterances per minute in recorded samples.

Infant vocalizations and the early diagnosis of severe hearing impairment.

To determine whether late onset of canonical babbling could be used as a criterion to determine risk of hearing impairment, we obtained vocalization samples longitudinally from 94 infants with normal hearing and 37 infants with severe to profound hearing impairment. Parents were instructed to report the onset of canonical babbling (the production of well-formed syllables such as "da," "na," "bee," "yaya"). Verification that the infants were producing canonical syllables was collected in laboratory audio recordings. Infants with normal hearing produced canonical vocalizations before 11 months of age (range, 3 to 10 months; mode, 7 months); infants who were deaf failed to produce canonical syllables until 11 months of age or older, often well into the third year of life (range, 11 to 49 months; mode, 24 months). The correlation between age at onset of the canonical stage and age at auditory amplification was 0.68, indicating that early identification and fitting of hearing aids is of significant benefit to infants learning language. The fact that there is no overlap in the distribution of the onset of canonical babbling between infants with normal hearing and infants with hearing impairment means that the failure of otherwise healthy infants to produce canonical syllables before 11 months of age should be considered a serious risk factor for hearing impairment and, when observed, should result in immediate referral for audiologic evaluation.

Perception of the stop/glide contrast in infancy.

Much of the published research in infant speech perception has emphasized how well infants have done with a number of speech contrasts, and have noted similarities in pattern of discrimination of adults and infants. Often it has been suggested that infants begin life with the ability to perceive any speech contrast, and that the process of acquiring a language involves inhibition of the ability to perceive contrasts not present in the target language. Indeed some studies have shown infants able to discriminate contrasts on which adults fail if the contrasts are not drawn from the native language of the adults. Other studies, however, have suggested that infants may not always be so perceptually capable. The present work focusses on the stop-glide contrast. The results are inconsistent with the prevalent view and with previously reported studies on the perception of the stop-glide contrast by infants. The results indicate that in a vigilance paradigm adapted for both infant and adult testing, infants perform poorly on the contrasts when compared with adults. Furthermore the pattern of relative perception observed in the adults on stimuli with long or short vowels is quite unlike that of the infants. It is concluded that much work remains in order to evaluate the relative performance of infants and adults in speech perception, since it appears that changes in experimental paradigm or particular stimulus parameters may affect outcomes in fundamental ways.