BabyHuBERT: Multilingual Self-Supervised Learning for Segmenting Speakers in Child-Centered Long-Form Recordings
BabyHuBERT is a self-supervised speech model trained on 13,000 hours of multilingual child-centered recordings from over 40 languages to classify speakers in long-form audio. It significantly outperforms existing models by handling the acoustic and linguistic challenges of children's speech, showing promise for automated analysis in early language development research.