Using a PCA-based dataset similarity measure to improve cross-corpus emotion recognition I Q1 X X TaggedPIngo Siegert a, *, Ronald B€ ock a,b , Andreas Wendemuth a,b TaggedP a Cognitive Systems Group, Faculty of Electrical Engineering and Information Technology, Otto von Guericke University, Magdeburg 39106, Germany b Center for Behavioral Brain Sciences, Magdeburg 39118, Germany Q2 X X Received 8 September 2016; received in revised form 22 August 2017; accepted 16 February 2018 Available online xxx TaggedPAbstract In emotion recognition from speech, huge amounts of training material are needed for the development of classification engines. As most current corpora do not supply enough material, a combination of different datasets is advisable. Unfortunately, data recording is done differently and various emotion elicitation and emotion annotation methods are used. Therefore, a combi- nation of corpora is usually not possible without further effort. The manuscript’s aim is to answer the question which corpora are similar enough to jointly be used as training material. A corpus similarity measure based on PCA-ranked features is presented and similar datasets are identified. To evaluate our method we used nine well-known benchmark corpora and automatically identified a sub-set of six most similar datasets. To test that the identified most similar six datasets influence the classification performance, we conducted several cross-corpora emotion recognition experiments comparing our identified six most similar datasets with other combinations. Our most similar sub-set outperforms all other combinations of corpora, the combination of all nine datasets as well as feature normalization techniques. Also influencing side-effects on the recognition rate were excluded. Finally, the pre- dictive power of our measure is shown: increasing similarity score, expressing decreasing similarity, result in decreasing recogni- tion rates. Thus, our similarity measure answers the question which corpora should be included into joint training. Ó 2018 Elsevier Ltd. All rights reserved. TaggedPKeywords: PCA; Dataset similarity; Cross-corpus emotion recognition; Automatic similarity scoring 1 1. Introduction 2 TaggedPIn the last decade, remarkable developments were obtained to improve speech based emotion recognition perfor- 3 mance in various ways (Anusuya and Katti, 2009; Koolagudi and Rao, 2012; Schuller et al., 2011a). This was 4 achieved by using sophisticated features and dedicated classifiers (Anagnostopoulos et al., 2012; Eyben et al., 2016; 5 Fern andez-Delgado et al., 2014). Also, considering the type of classification engines, we nowadays have the oppor- 6 tunity to utilize a variety of methods like Deep Neural Networks (Stuhlsatz et al., 2011; Li et al., 2013), Hidden I This paper has been recommended for acceptance by Roger K. Moore. * Corresponding author.D102X X E-mail address: ingo.siegert@ovgu.de (I. Siegert). http://dx.doi.org/10.1016/j.csl.2018.02.002 0885-2308/ 2018 Elsevier Ltd. All rights reserved. Available online at www.sciencedirect.com Computer Speech & Language xxx (2018) xxx-xxx www.elsevier.com/locate/csl ARTICLE IN PRESS JID: YCSLA [m3+;February 28, 2018;7:10] Please cite this article as: I. Siegert et al., Using a PCA-based dataset similarity measure to improve cross-corpus emotion recognition, Computer Speech & Language (2018), http://dx.doi.org/10.1016/j.csl.2018.02.002