Researchers at UC Berkeley are doing something similar called SSDM (Stuttering Speech Disfluency Model) to teach algorithms how to recognize disfluency : https://berkeley-speech-group.github.io/SSDM/ Also Dr Ghosh and her team at Cornell are developing StutterCut which is a segmentation tool that marks exactly when and where in your audio the repetition or block happens down to milliseconds : https://arxiv.org/html/2508.02255v1