StutterCut is a semi-supervised graph-based framework for segmenting speech dysfluencies without requiring strong labels or transcriptions. It partitions speech embeddings using a soft-constrained clustering method based on Normalised Cut (N-Cut), enhanced by uncertainty-aware classifier guidance.
📢 Coming Soon
The code and dataset will be made publicly available shortly. Please stay tuned!
✅ Our paper on StutterCut has been accepted at Interspeech 2025!
We’re excited to share our work with the speech community.
If you use our method or dataset, please cite:
@inproceedings{ghosh2025stuttercut, title = {StutterCut: Uncertainty-Guided Normalised Cut for Dysfluency Segmentation}, author = {Suhita Ghosh and Melanie Jouaiti and Jan-Ole Perschewski and Sebastian Stober}, booktitle = {Proceedings of Interspeech 2025}, year = {2025} }
To receive updates when the code and data are released, consider watching this repository or opening an issue.
The code will be released under the MIT License.