Stephen McIntosh

A DTW-based framework using self-supervised representations outperforms human raters in assessing L2 phonetic accuracy and rhythm, challenging traditional assessment methods.

Phonological features can be extracted from self-supervised speech models in under a minute, achieving state-of-the-art performance in phone segmentation and recognition.

Speech Playground transforms speech analysis by enabling seamless integration of deep learning representations with interactive visualization tools.

Turns out, you can measure how well speech models capture subtle prosodic differences like stress and tone using just a few unlabeled examples.