Collecting child speech data is particularly challenging, requiring a controlled and developmentally sensitive approach to account for the unique characteristics of child speech.

With over one million children affected by speech and language impairments annually, early identification and intervention are crucial for better outcomes.

The high-quality dataset resulting from this research will enhance the capability of AI to assist clinicians in diagnosing speech impairments, especially in areas with limited access to specialists.

This innovation allows speech-language pathologists and educators to identify speech-language concerns earlier, improving support for children in need.

Ultimately, AI-powered systems will play a vital role in helping professionals address speech-language issues more effectively.

Speights and her team faced a dilemma: they needed automated tools to efficiently collect data, but lacked the large datasets necessary to train such tools.

Marisha Speights, an assistant professor at Northwestern University, has developed a data pipeline aimed at training AI tools specifically for childhood speech screening.

She presented her innovative research at the joint 188th Meeting of the Acoustical Society of America and the 25th International Congress on Acoustics held from May 18 to May 23, 2025.

To create a comprehensive dataset, Speights and her team constructed a computational pipeline that transforms raw child speech data into a format suitable for AI training, ensuring the quality and accuracy of the recordings.

However, they faced significant challenges in processing and annotating thousands of speech samples without the automated tools they are developing.

The need for such tools is underscored by the fact that current AI speech recognition systems are primarily trained on adult speech, rendering them ineffective for diagnosing children's speech impairments.