Self-Supervised learning with cross-modal transformers for emotion recognition

TL;DR

This paper introduces a self-supervised cross-modal transformer model trained on audio, visual, and text data to improve emotion recognition, demonstrating a 3% performance boost on the CMU-MOSEI dataset.

Contribution

It extends self-supervised learning to multi-modal emotion recognition using a transformer trained with masked modeling on multiple data modalities.

Findings

Improved emotion recognition accuracy by up to 3%.

Effective multi-modal representation learning with self-supervised pre-training.

Demonstrated benefits of cross-modal transformers in emotion recognition.

Abstract

Emotion recognition is a challenging task due to limited availability of in-the-wild labeled datasets. Self-supervised learning has shown improvements on tasks with limited labeled datasets in domains like speech and natural language. Models such as BERT learn to incorporate context in word embeddings, which translates to improved performance in downstream tasks like question answering. In this work, we extend self-supervised training to multi-modal applications. We learn multi-modal representations using a transformer trained on the masked language modeling task with audio, visual and text features. This model is fine-tuned on the downstream task of emotion recognition. Our results on the CMU-MOSEI dataset show that this pre-training technique can improve the emotion recognition performance by up to 3% compared to the baseline.