Contrastive Unsupervised Learning for Speech Emotion Recognition

TL;DR

This paper demonstrates that contrastive predictive coding (CPC) can learn effective speech emotion representations from unlabeled data, significantly improving emotion recognition accuracy on multiple datasets.

Contribution

It introduces the application of CPC for unsupervised speech emotion recognition, achieving state-of-the-art results without labeled data.

Findings

Achieved state-of-the-art CCC on IEMOCAP for all emotion primitives.

Significant performance improvements on MSP-Podcast dataset.

Unsupervised learning enhances SER performance without labeled datasets.

Abstract

Speech emotion recognition (SER) is a key technology to enable more natural human-machine communication. However, SER has long suffered from a lack of public large-scale labeled datasets. To circumvent this problem, we investigate how unsupervised representation learning on unlabeled datasets can benefit SER. We show that the contrastive predictive coding (CPC) method can learn salient representations from unlabeled datasets, which improves emotion recognition performance. In our experiments, this method achieved state-of-the-art concordance correlation coefficient (CCC) performance for all emotion primitives (activation, valence, and dominance) on IEMOCAP. Additionally, on the MSP- Podcast dataset, our method obtained considerable performance improvements compared to baselines.