Cross-Corpus Acoustic Emotion Recognition: Variances and Strategies

Schuller, Björn; Vlasenko, Bogdan; Eyben, Florian; Wöllmer, Martin; Stuhlsatz, André; Wendemuth, Andreas; Rigoll, Gerhard

doi:10.1109/t-affc.2010.8

Cited by 328 publications

(288 citation statements)

References 60 publications

Supporting

Mentioning

253

Contrasting

Order By: Relevance

“…We quantified this fact by benchmarks reported on nine frequently used datasets in [7,8]. To raise these and advance speech-based emotion recognition systems' performance, we introduce a Generalized Discriminant Analysis (GerDA) [13] that is a recently proposed machine learning tool based on Deep Neural Networks (DNNs) for discriminative feature extraction from arbitrary distributed raw data.…”

Section: Introductionmentioning

confidence: 99%

Deep neural networks for acoustic emotion recognition: Raising the benchmarks

Stuhlsatz

Meyer

Eyben

et al. 2011

2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

242

118

View full text Add to dashboard Cite

Deep Neural Networks (DNNs) denote multilayer artificial neural networks with more than one hidden layer and millions of free parameters. We propose a Generalized Discriminant Analysis (GerDA) based on DNNs to learn discriminative features of low dimension optimized with respect to a fast classification from a large set of acoustic features for emotion recognition. On nine frequently used emotional speech corpora, we compare the performance of GerDA features and their subsequent linear classification with previously reported benchmarks obtained using the same set of acoustic features classified by Support Vector Machines (SVMs). Our results impressively show that low-dimensional GerDA features capture hidden information from the acoustic features leading to a significantly raised unweighted average recall and considerably raised weighted average recall.

show abstract

Section: Introductionmentioning

confidence: 99%

Deep neural networks for acoustic emotion recognition: Raising the benchmarks

Stuhlsatz

Meyer

Eyben

et al. 2011

2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Self Cite

242

118

View full text Add to dashboard Cite

show abstract

“…Most studies tend to overestimation in this respect: Acted data are often used rather than spontaneous data, results are reported on preselected prototypical data, and true speaker disjunctive partitioning is still less common than simple cross-validation. Even speaker disjunctive evaluation can give only little insight into the generalization ability of today's emotion recognition engines since training and test data as used for system development usually tend to be similar as far as recording conditions, noise overlay, language, and types of emotions are concerned [2]. For example, if a system builds upon a classifier using features extracted from adults' speech corpora to identify children's emotional state, its performance can be expected as very low.…”

Section: Related Workmentioning

confidence: 99%

“…For the comparability with FAU AEC, we additionally map the diverse emotion groups onto the valence axis in the dimensional emotion model. The mapping defined in [2] for cross-corpus experiments is used to generate labels for binary valence from the emotion categories in order to generate a unified set of labels. This mapping is given in Table I.…”

Section: Databasesmentioning

confidence: 99%

“…For that reason, much of the actual effort in deploying systems of speech emotion recognition goes into the design of an appropriate representation of the data in order to support promising classification performance. At present, several emotional speech corpora exist, but they are typically highly dissimilar in terms of spoken language, type of emotion such as acted, elicited, or naturalistic, or type of labelling scheme such as categorical or dimensional [2]. When labelling emotional corpora, even worse, there is no certain ground truth but a subjective ambiguous 'gold standard' as given by majority voting of several human raters which may be in considerable disagreement.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Sparse Autoencoder-Based Feature Transfer Learning for Speech Emotion Recognition

Deng

Zhang

Marchi

et al. 2013

2013 Humaine Association Conference on Affective Computing and Intelligent Interaction

Self Cite

309

182

View full text Add to dashboard Cite

Abstract-In speech emotion recognition, training and test data used for system development usually tend to fit each other perfectly, but further 'similar' data may be available. Transfer learning helps to exploit such similar data for training despite the inherent dissimilarities in order to boost a recogniser's performance. In this context, this paper presents a sparse autoencoder method for feature transfer learning for speech emotion recognition. In our proposed method, a common emotion-specific mapping rule is learnt from a small set of labelled data in a target domain. Then, newly reconstructed data are obtained by applying this rule on the emotion-specific data in a different domain. The experimental results evaluated on six standard databases show that our approach significantly improves the performance relative to learning each source domain independently.

show abstract

“…Such features include, but are not limited to, speech and its content, prosodic and paralinguistic features, eye gaze, facial expressions, body movements, or more advanced interpretations of such features such as the affective state, personality, mood or intentions of the user (e.g. [8,14,15]). …”

Section: Analysis Of Natural Interactionsmentioning

confidence: 99%