Does Visual Self-Supervision Improve Learning of Speech Representations for Emotion Recognition?

Abstract

Self-supervised learning has attracted plenty of recent research interest. However, most works are typically unimodal and there has been limited work that studies the interaction between audio and visual modalities for self-supervised learning. This work (1) investigates visual self-supervision via face reconstruction to guide the learning of audio representations; (2) proposes two audio-only self-supervision approaches for speech representation learning; (3) shows that a multi-task combination of the proposed visual and audio self-supervision is beneficial for learning richer features that are more robust in noisy conditions; (4) shows that self-supervised pretraining leads to a superior weight initialization, which is especially useful to prevent overfitting and lead to faster model convergence on smaller sized datasets. We evaluate our audio representations for emotion and speech recognition, achieving state of the art performance for both problems. Our results demonstrate the potential of visual self-supervision for audio feature learning and suggest that joint visual and audio self-supervision leads to more informative speech representations.

Task Programming: Learning Data Efficient Behavior Representations

“…We jointly optimize a family of self-supervised tasks in an encoderdecoder setup, making this work an example of multitask self-supervised learning. Multi-task self-supervised learning has been applied to other domains such as visual data [11,24], accelerometer recordings [35], audio [34] and multi-modal inputs [37,30]. Generally in each of these domains, tasks are defined ahead of time, as is the case for tasks such as frame reconstruction, colorization, finding relative position of image patches, mapping videos to optimal flow, and video-audio alignment.…”