Improving Cross-Corpus Speech Emotion Recognition with Adversarial Discriminative Domain Generalization (ADDoG)
Улучшение распознавания эмоций в речи между корпусами с помощью состязательного дискриминативного обобщения по доменам (ADDoG)
2019-05-14
SCID: 54.1/uj44h7rw
Discuss with AI
adversarial discriminative domain generalizationcross-corpus speech emotion recognitioncross-dataset generalizationdomain generalizationin-the-wild speech data
Figures from the paper
Abstract (AI)
Automatic speech emotion recognition provides computers with critical context to enable user understanding. While methods trained and tested within the same dataset have been shown successful, they often fail when applied to unseen datasets. To address this, recent work has focused on adversarial methods to find more generalized representations of emotional speech. However, many of these methods have issues converging, and only involve datasets collected in laboratory conditions. In this paper, we introduce Adversarial Discriminative Domain Generalization (ADDoG), which follows an easier to train "meet in the middle" approach. The model iteratively moves representations learned for each dataset closer to one another, improving cross-dataset generalization. We also introduce Multiclass ADDoG, or MADDoG, which is able to extend the proposed method to more than two datasets, simultaneously. Our results show consistent convergence for the introduced methods, with significantly improved results when not using labels from the target dataset. We also show how, in most cases, ADDoG and MADDoG can be used to improve upon baseline state-of-the-art methods when target dataset labels are added and in-the-wild data are considered. Even though our experiments focus on cross-corpus speech emotion, these methods could be used to remove unwanted factors of variation in other settings.
Key Findings
1
ADDoG and MADDoG generally improve state-of-the-art baselines when target labels are available and when evaluating on in-the-wild data.
2
ADDoG and MADDoG show consistent convergence, addressing training-convergence issues reported for prior adversarial domain-generalization methods.
3
ADDoG introduces a “meet in the middle” adversarial approach that iteratively aligns representations learned from different speech emotion datasets.
4
Multiclass ADDoG (MADDoG) extends the method to align representations from more than two datasets simultaneously.
5
The proposed methods significantly improve cross-dataset emotion recognition without using labels from the target dataset.
Research Object
cross-corpus speech emotion recognition across laboratory and in-the-wild speech datasets
Research Subject
adversarial domain-generalized representation learning and cross-dataset generalization of emotional speech, including convergence and performance with and without target-dataset labels
Publication Details
Publication Date
2019-05-14
Journal
Publisher
ISSN
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest