Speaker-Invariant Domain-Adversarial Neural Networks

Source codes for our ICMI'20 paper: Speaker-Invariant Adversarial Domain Adaptation for Emotion Recognition

Source Listing

experiment_a and experiment_v

Source codes of detecting arousal and valence values. Implemented models include baseline, domain-adversarial neural networks (DANN), and speaker-invariant DANN (SIDANN)

preprocess_iemocap and preprocess_msp_improv

Feature extraction codes for the two datasets. You need to download ResNet-152 model and VGGish model to /models and /AUDIOSET for feature extraction.