Reproduce “More Similar than Dissimilar: Modeling Annotators for Cross-Corpus Speech Emotion Recognition” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
James Tavernor, Emily Mower Provost
This research introduces a framework that models human annotators for cross-corpus speech emotion recognition, enabling personalized predictions for new, unseen annotators. It achieves this by identifying a similar pre-trained annotator model using a minimal amount of labeled enrollment data, enhancing both individual and aggregated emotion prediction without extensive retraining.
Key insights
- Modeling individual annotator perceptions and leveraging inter-annotator similarities can significantly improve both individual and aggregate emotion prediction in SER, especially in cross-corpus settings.
- A small enrollment set of 15-20 annotations per target annotator is sufficient to accurately identify a similar source annotator and achieve robust personalized performance.
- The selection of similar source annotators for a given target annotator is consistent across different cross-validation folds and random seeds, suggesting robust identification of genuine perceptual commonalities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.