EMO-Debias: Benchmarking Gender Debiasing Techniques in Multi-Label Speech Emotion Recognition
The episode discusses 'EMO-Debias,' a paper benchmarking gender debiasing techniques for multi-label speech emotion recognition. Researchers tested thirteen methods across two datasets and found that some single-label fixes fail in multi-label settings. GADRO and LfF proved most robust, emphasizing the need for better data collection and careful selection of fairness methods for building trustworthy AI.