Conventional semi-supervised learning (SSL) encounters challenges in effectively addressing issues associated with long-tail datasets, primarily stemming from imbalances within a dataset. Existing ...

In this paper we present our findings on how representation learning on large unlabeled speech corpora can be beneficially utilized for speech emotion recognition (SER). Prior work on representation ...