Context Modeling for Cross-Corpus Dimensional Acoustic Emotion Recognition: Challenges and Mixup
Küçük Resim Yok
Tarih
2018
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Springer Verlag
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Recently, focus of research in the field of affective computing was shifted to spontaneous interactions and time-continuous annotations. Such data enlarge the possibility for real-world emotion recognition in the wild, but also introduce new challenges. Affective computing is a research area, where data collection is not a trivial and cheap task; therefore it would be rational to use all the data available. However, due to the subjective nature of emotions, differences in cultural and linguistic features as well as environmental conditions, combining affective speech data is not a straightforward process. In this paper, we analyze difficulties of automatic emotion recognition in time-continuous, dimensional scenario using data from RECOLA, SEMAINE and CreativeIT databases. We propose to employ a simple but effective strategy called “mixup” to overcome the gap in feature-target and target-target covariance structures across corpora. We showcase the performance of our system in three different cross-corpus experimental setups: single-corpus training, two-corpora training and training on augmented (mixed up) data. Findings show that the prediction behavior of trained models heavily depends on the covariance structure of the training corpus, and mixup is very effective in improving cross-corpus acoustic emotion recognition performance of context dependent LSTM models. © 2018, Springer Nature Switzerland AG.
Açıklama
20th International Conference on Speech and Computer, SPECOM 2018 -- 18 September 2018 through 22 September 2018 -- Leipzig -- 218179
Anahtar Kelimeler
Cross-corpus emotion recognition, Data augmentation, Time-continuous emotion recognition
Kaynak
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
WoS Q Değeri
Scopus Q Değeri
Q3
Cilt
11096 LNAI
