ExELang Legacy Ecosystem
The ExELang Legacy Ecosystem is a set of agreements designed to ensure that the work produced during the ExELang project continues to benefit the research community. It brings together three independent structures: the ExELang Consortium, the EL1000 Consortium, and ELSI.
ExELang Consortium Information Booklet
The ExELang Consortium Information Booklet describes how the ExELang Consortium works: who can do what with the data, under what conditions, and who is responsible for what. It is intended for corpus collectors, institutional review boards, legal and compliance departments, and scientific archives.
Use Case: “I want to understand how the ExELang Consortium works and what my rights are as a corpus collector.”
The ExELang Consortium Information Booklet explains the governance framework, data access rules, and publication policies of the consortium.
ELSI
ELSI (ExELang Legacy Support Interface) is a web-based platform that helps researchers manage standardized datasets, retrieve machine-learning annotations, and extract metrics from child-centered, long-form recordings. In June 2026, the platform was only available to test users, and given the end of ExELang funding, we are not able to provide free access nor ensure its maintenance in the future.
Use Case: “I want to analyze child language recordings”
Follow the ELSI guide to process and analyze your audio data effectively.
EL1000 Consortium
EL1000 is a research consortium for large-scale, cross-cultural research on children’s everyday language environments. Rather than sharing raw audio, it uses automated speech analysis pipelines to generate deidentified derived metrics — enabling standardized, cross-corpus analyses while respecting data sovereignty and consent constraints.
ChildProject
Project management and data organization for child language research.
Diarization Simulation
A Python package for testing how classification errors in automated speaker diarization affect your research findings.
Use Case: “I want to check whether the patterns I found in my data could be artifacts of diarization errors rather than real effects”
Diarization algorithms like LENA and VTC can misclassify speakers — for example, confusing child and female adult voices — and these errors can silently distort downstream analyses. Diarization Simulation lets you provide your ground truth vocalization data and simulate what those algorithms would actually measure, so you can assess how sensitive your findings are to classification errors.
VTC2 Documentation
Voice Type Classification tool for audio analysis.
Use Case: “I need to classify voices in my audio”
VTC2 helps you identify and classify different voice types in recordings.
BabyHubert
Advanced machine learning model for infant speech analysis.
Speech Maturity Dataset
A large, publicly available dataset of child vocalizations for studying early speech development across languages and communities.
Use Case: “I want to train or benchmark a model on child speech across diverse languages and communities”
The Speech Maturity Dataset (SMD) contains over 258,000 manually labelled audio clips of vocalizations from children aged 2 months to 6 years, drawn from 398 children across 14 communities and 25+ languages — spanning industrialized societies to farmer-forager communities.
Benchmarking Dataset Factory
A pipeline for building custom benchmarking datasets for supervised machine learning models, using human-annotated long-form recording data.
Use Case: “I want to train or evaluate a speech model and need a clean, standardised dataset drawn from human-annotated child recordings”
The Benchmarking Dataset Factory lets you combine annotation data across multiple long-form recording corpora into a single, ready-to-use benchmarking dataset. It supports four annotation types — voice type classification, vocalization/speech maturity, addressee, and transcription — and handles the messy work of harmonising column names across datasets, filtering recordings, and splitting data into train/test/validation sets. Datasets are built in the ChildProject format, making them compatible with the rest of the ExELang toolchain.
Tutorial for Navigating Ethical Dilemmas in Research
A practical guide for researchers facing ethical decisions in early language intervention studies.
Use Case: “I’m designing a study involving language interventions with young children and need help thinking through the ethical trade-offs”
This tutorial walks researchers through common ethical dilemmas that arise in early language intervention research — such as balancing scientific goals with participant wellbeing and community impact — and provides a structured framework for reasoning through them.
Data Protection Handbook for Long-Form Recording Research
A global guide to data protection laws for researchers collecting long-form audio recordings.
Use Case: “I’m collecting recordings in a new country and need to understand what data protection rules apply to my study”
Long-form recordings capture sensitive, unfiltered aspects of participants’ daily lives and are subject to varying legal frameworks depending on where research takes place. This handbook helps researchers navigate data protection laws across different national contexts, so they can collect and manage recordings compliantly and responsibly.
Long-Form Recordings in Low- and Middle-Income Countries
Recommendations for conducting respectful long-form recording research in LMICs.
Use Case: “I want to expand my research to communities outside high-income countries and need guidance on ethical best practices”
As long-form recording research expands beyond Western, high-income settings, it brings new ethical challenges around consent, community representation, and power dynamics. This paper identifies four key ethical challenges specific to research in low- and middle-income countries and offers concrete recommendations to address them.
Automated Feedback Pipeline for Caregiver-Child Conversations
An end-to-end pipeline for automatically analysing short caregiver-child audio recordings and generating personalised feedback.
Use Case: “I’m running an early childhood intervention and want to give caregivers automated, personalised feedback on their conversations with their child — without manual annotation”
This pipeline processes 1–10 minute smartphone recordings of caregiver-child conversations, automatically analyses them using open-source tools and AWS, and generates personalised feedback for caregivers. It is designed to be cheap (around $0.20 per hour of audio), reproducible, and scalable — making it practical for large-scale intervention programmes. The pipeline is already operational in a real-world study in Uruguay, where audio files are collected via a chatbot, and could be adapted for other early childhood or caregiver-support contexts.
Teaching Materials
A comprehensive, open-access introduction to long-form recording research — from study design to data analysis.
Use Case: “I’m new to long-form recordings and want a structured resource to get up to speed”
Long-form recordings: From A to Z is a free, open-access book covering the full research pipeline: what long-form recordings are, how to collect and manage them, how to process and annotate them, and how to analyse the data. It is designed for researchers at any level, with or without a technical background.
Need Help?
Can't find what you're looking for? Contact our team or check out our GitHub repositories.