Overview of LENA development and surprising findings, both confirming and expanding on Hart and Risley’s ground breaking work.

This paper describes the multiphase LENA Natural Language Study, an ongoing data collection effort designed to investigate the language environment of infants and toddlers. Data collected contributes to product development and normative information for use with the LENA System and child development research. Phase I study participants were representative of the U.S. Census with respect to mothers’ attained education and consisted of 329 normally developing infants and toddlers from monolingual English-speaking households living in the Denver-metro area. Participants provided daylong audio recordings of their natural language environment once a month, and certified speech language pathologists assessed participant language ability independently through standardized assessments. A subset of 80 phase I participants has continued to provide monthly recordings in phase II of the study. The normative database described herein contains over 32,000 hours of spontaneous speech data. This paper describes how normative information was derived for the Adult Word Count estimates (AWC; adult words spoken per day), Conversational Turns estimates (CT; adult-child alternations per day), and Child Vocalization frequency estimates (CV; words, babbles, and “protophones” or pre-speech communicative sounds) that are reported in the LENA System.

The LENA language environment analysis system was designed to estimate adult and key child interactions in natural home environments. Contrary to controlled clinical research environments, the speech used by the participants in this study was real, unrehearsed, and representative of each child’s typical daily language environment. In this paper, we describe the audio processing system in terms of information low, feature extraction, and segmentation identification. We also reveal the audio specifications that were either met or exceeded during the development and design of the LENA Digital Language Processor (DLP).

The LENA Digital Language Processor (DLP) records real-time audio data that is transferred to a computer for processing by LENA language environment software V3.1.0. Ultimately an Interpreted Time Segments (ITS) file is created that summarizes the processing results for the data contained within the original audio file. LENA data analyses have focused primarily on estimating Adult Word Counts (AWC), Child Vocalizations (CV), and Conversational Turns (CT) between adult and key child. however, the data processing reveals numerous other factors that have not yet been studied by researchers at LENA Foundation. In this paper, we reveal the general content of the ITS file. We also provide a detailed description of each component specific to the ITS file and implications with regard to potential research and analyses.

The development of the LENA language environment analysis software V3.1.0 as well as quantifying its accuracy and reliability was dependent on the accurate and reliable transcription of the audio recording files by professional transcribers. In this report, we discuss the analytical procedure designed and implemented by LENA’s professional transcription team to identify and code speakers. We also reveal the degree to which inter-rater reliability was achieved between the criterion rater and four to seven secondary raters in terms of agreement for segment classification and adult word counts. Significant accuracy in the transcriptions was critical to train the processing models used to identify segments in audio recordings automatically and subsequently to test the accuracy and reliability of the segmentation process.

As part of the development of the LENA System, a team of LENA speech language pathologists, linguists, statisticians, and other researchers created the LENA Developmental Snapshot. This 52-item survey was designed to assess expressive and receptive language skills in children 2 months to 36 months of age and to estimate developmental age as a function of chronological age. In this report, we describe the development of the Developmental Snapshot. Initial respondents were drawn largely from the LENA Normative Study participant pool, and their scores were validated based on correlations with pre-existing standardized language and cognitive assessments. Participants in a longitudinal study completed the Developmental Snapshot at approximately monthly intervals to evaluate test-retest reliability.

This report describes the development of the LENA Automatic Vocalization Assessment (AVA) software. AVA software was designed to provide both parents and professionals with automatically generated information about the expressive language development of children ages 2 months to 48 months. Expressive language estimates are produced based on 12- to 16-hour audio recordings collected in the natural home environment using the LENA language environment analysis system. AVA software uses automatic speech recognition technology to categorize and quantify the sounds in child vocalizations (e.g., protophones and phonemes). These quantitative acoustic information data (expressed as “phone” and “biphone” frequencies) are reduced to principal components which are applied as input for age-based multiple linear regression models. The AVA software utilizes these regression models to generate information about expressive language development as standard scores, developmental age estimates, and estimated mean length of utterance (EMLU). AVA expressive language estimates demonstrate statistical reliability and validity comparable to standard expressive language assessments commonly administered by speech language pathologists.

The LENA language environment analysis system is a powerful new tool that can help pediatricians, speech language pathologists, and audiologists improve their diagnosis and treatment of language delay and disorders, and increase parent involvement and satisfaction. Broadly speaking, the uses of LENA in clinical practice fall into four categories: 1) assessment and diagnosis, 2) monitoring fidelity of treatment, 3) audio environment analysis, 4) enhanced treatment through in-home feedback.

Autism Spectrum Disorders (ASD) are characterized by qualitative impairments in social interaction and communication as well as restricted and repetitive behaviors. The presence or absence of these behavioral signs underlies the criteria that clinicians use to assess children for ASD. However, the extensive training and expertise required to diagnose ASD and the sometimes incomplete nature of available information can limit the efficiency and reliability of screening using traditional indicators, especially at younger ages. Going beyond established diagnostic criteria, researchers have reported atypicality in the vocal production of children with ASD for features such as duration, pitch, and rhythm. Such anomalies potentially carry important diagnostic information, but on a practical level the means to explore this possibility in greater detail have been lacking. In particular, the identification of vocal features characteristic of ASD has been limited by the need to rely on resource-intensive expert judgment and the difficulty of obtaining, processing, and interpreting representative audio samples of sufficient quality and quantity. Here we report on the development and performance of a fully automatic and objective method that utilizes recent advances in technology to collect child vocalizations in large volume and evaluate discriminative vocal characteristics that could be used to help identify children at risk for ASD.