Continuous Multimodal Emotion Prediction Based on Long Short Term Memory Recurrent Neural Network
The DOI matches its Crossref record; 33 of the 34 references it cites were found in a registry or an index.
Abstract
The continuous dimensional emotion can depict subtlety and complexity of emotional change, which is an inherently challenging problem with growing attention. This paper presents our automatic prediction of dimensional emotional state for Audio-Visual Emotion Challenge (AVEC 2017), which uses multi-features and fusion across all available modalities. Besides the baseline features provided by the organizers, we also extract other acoustic audio feature sets, appearance features and deep visual features as complementary features. Each type of feature is trained using Long Short-Term Memory Recurrent Neutral Network (LSTM-RNN) for every dimensional emotion prediction separately considering annotation delay and temporal pooling. To overcome overfitting problem, robust models are chosen carefully for individual model. Finally, multimodal emotion fusion is achieved by utilizing Support Vector Regression (SVR) with the estimates from different feature sets in decision level fusion. The experimental results indicate that our extracted features are beneficial to performance improvement and our system design achieves very promising results with Concordant Correlation Coefficient (CCC), which outperform the baseline system on the testing set for arousal of 0.599 vs 0.375 (baseline) and for valence of 0.721 vs 0.466 and for liking 0.295 vs 0.246.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Furthermore, Huang et al. [86] adopted Long Short-Term Memory (LSTM) networks to create affect prediction models that fully leverage affective information from each modality. They formulated prediction models for various feature types within each modality and used Support Vector Regression (SVR) as the fusion model in the decision layer.”
From A Survey of Deep Learning-Based Multimodal Emotion Recognition: Speech, Text, and Face (2023) · cites this paper as [86]
Citations
Cites
Histograms of Oriented Gradients for Human Detection
Dalal, N., & Triggs, B. (2005) · 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) · doi:10.1109/cvpr.2005.177 · Cited by 33,584 · cited 1×
“HOG [28] features describe the distribution information of intensity gradients or edge directions.”
BackgroundADADELTA: An Adaptive Learning Rate Method
Zeiler, M. D. (2012) · arXiv · doi:10.48550/arxiv.1212.5701 · Cited by 5,322
Opensmile: the munich versatile and fast open-source audio feature extractor
Eyben, F., Wöllmer, M., & Schuller, B. (2010) · Proceedings of the 18th ACM international conference on Multimedia · doi:10.1145/1873951.1874246 · Cited by 2,642 · cited 1×
“The extraction of the LLDs and the computation of the functionals are done using the openSMILE toolkit [25].”
MethodThe Geneva Minimalistic Acoustic Parameter Set (GeMAPS) for Voice Research and Affective Computing
Eyben, F., Scherer, K. R., Schuller, B. W., Sundberg, J., Andre, E., Busso, C., Devillers, L. Y., Epps, J., Laukka, P., Narayanan, S. S., et al. (2016) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2015.2457417 · Cited by 1,818 · cited 1×
“For the audio signals, AVEC organizers provide the extended Geneva Minimalistic Acoustic Parameter Set (eGeMAPS) [13] as baseline features.”
MethodAMIGOS: A Dataset for Affect, Personality and Mood Research on Individuals and Groups
Miranda-Correa, J. A., Abadi, M. K., Sebe, N., & Patras, I. (2021) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2018.2884461 · Cited by 709
AVEC 2013: the continuous audio/visual emotion and depression recognition challenge
Valstar, M., Schuller, B., Smith, K., Eyben, F., Jiang, B., Bilakhia, S., Schnieder, S., Cowie, R., & Pantic, M. (2013) · Proceedings of the 3rd ACM international workshop on Audio/visual emotion challenge · doi:10.1145/2512530.2512533 · Cited by 581 · cited 1×
“The depression recognition sub-challenge is added since AVEC 2013 [7], which is related to emotion recognition well.”
BackgroundThe INTERSPEECH 2010 paralinguistic challenge
Schuller, B., Steidl, S., Batliner, A., Burkhardt, F., Devillers, L., Müller, C., & Narayanan, S. S. (2010) · Interspeech 2010 · doi:10.21437/interspeech.2010-739 · Cited by 571 · cited 1×
“Specifically, we extract the baseline audio feature set of INTERSPEECH 2010 Paralinguistic Challenge (IS10) [26] which can better reflect a border coverage of paralinguistic information assessment.”
MethodAVEC 2016: Depression, Mood, and Emotion Recognition Workshop and Challenge
Valstar, M., Gratch, J., Schuller, B., Ringeval, F., Lalanne, D., Torres Torres, M., Scherer, S., Stratou, G., Cowie, R., & Pantic, M. (2016) · Proceedings of the 6th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2988257.2988258 · Cited by 503 · cited 4×
“The physiological modality is added in AVEC 2015 [9] and AVEC 2016 [10], but doesn’t appear in AVEC 2017 [11].”
BackgroundMethodInfluentialTao, J., & Tan, T. (2005) · Lecture Notes in Computer Science · doi:10.1007/11573548_125 · Cited by 494 · cited 2×
“In SVM, the parameter C varies from range [1, 100]”
BackgroundSelf-citationAVEC 2014: 3D Dimensional Affect and Depression Recognition Challenge
Valstar, M., Schuller, B., Smith, K., Almaev, T., Eyben, F., Krajewski, J., Cowie, R., & Pantic, M. (2014) · Proceedings of the 4th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2661806.2661807 · Cited by 472 · cited 1×
“Michel et al. [8] calculate the average ratings from all raters to minimize the label noise.”
BackgroundAutomatic, Dimensional and Continuous Emotion Recognition
Gunes, H., & Pantic, M. (2010) · International Journal of Synthetic Emotions · doi:10.4018/jse.2010101605 · Cited by 426 · cited 2×
“Moreover, there is a shift from discrete emotion model to dimensional emotion model which describes emotion in a continuous multi-dimensional emotion space [3][4] and thus can model subtle and complicated emotional behaviors.”
BackgroundGunes, H., & Schuller, B. (2013) · Image and Vision Computing · doi:10.1016/j.imavis.2012.06.016 · Cited by 360 · cited 1×
“Moreover, there is a shift from discrete emotion model to dimensional emotion model which describes emotion in a continuous multi-dimensional emotion space [3][4] and thus can model subtle and complicated emotional behaviors.”
BackgroundAVEC 2017: Real-life Depression, and Affect Recognition Workshop and Challenge
Ringeval, F., Schuller, B., Valstar, M., Gratch, J., Cowie, R., Scherer, S., Mozgai, S., Cummins, N., Schmitt, M., & Pantic, M. (2017) · Proceedings of the 7th Annual Workshop on Audio/Visual Emotion Challenge · doi:10.1145/3133944.3133953 · Cited by 332 · cited 6×
“The factor of annotation delay is processed in many emotion recognition systems of AVEC [11][23][32] to improve the prediction performance.”
BackgroundMethodInfluentialLSTM-Modeling of continuous emotions in an audiovisual affect recognition framework
Wöllmer, M., Kaiser, M., Eyben, F., Schuller, B., & Rigoll, G. (2013) · Image and Vision Computing · doi:10.1016/j.imavis.2012.03.001 · Cited by 288 · cited 1×
“Wöllmer et al. [21] utilize LSTM-RNN to perform regression analysis on arousal and valence dimension, which improves recognition performance significantly.”
MethodAVEC 2011–The First International Audio/Visual Emotion Challenge
Schuller, B., Valstar, M., Eyben, F., McKeown, G., Cowie, R., & Pantic, M. (2011) · Lecture Notes in Computer Science · doi:10.1007/978-3-642-24571-8_53 · Cited by 271 · cited 3×
“In random forest, the number of trees varies from range [50, 500] and the maximum depth of the tree varies from range [5, 10].”
BackgroundAVEC 2012: the continuous audio/visual emotion challenge
Schuller, B., Valster, M., Eyben, F., Cowie, R., & Pantic, M. (2012) · Proceedings of the 14th ACM international conference on Multimodal interaction · doi:10.1145/2388676.2388776 · Cited by 226 · cited 1×
“The Audio-Visual Emotion Challenge (AVEC) [5-11], an annual challenge since 2011, aims at promoting the development of multimedia processing and machine learning methods for automatic continuous emotion recognition in the wild.”
BackgroundPrediction of asynchronous dimensional emotion ratings from audiovisual and physiological data
Ringeval, F., Eyben, F., Kroupi, E., Yuce, A., Thiran, J.-P., Ebrahimi, T., Lalanne, D., & Schuller, B. (2015) · Pattern Recognition Letters · doi:10.1016/j.patrec.2014.11.007 · Cited by 202 · cited 2×
“Ringeval et al. [32] add average window to the labels and features to smooth the labels and decrease the noise.”
BackgroundCombining Multiple Kernel Methods on Riemannian Manifold for Emotion Recognition in the Wild
Liu, M., Wang, R., Li, S., Shan, S., Huang, Z., & Chen, X. (2014) · Proceedings of the 16th International Conference on Multimodal Interaction · doi:10.1145/2663204.2666274 · Cited by 202 · cited 1×
“The effective Deep CNN model is trained using 110,000 face images from 1032 people in all, inspired by Liu’s work [29].”
MethodLocal Gabor Binary Patterns from Three Orthogonal Planes for Automatic Facial Expression Recognition
Almaev, T. R., & Valstar, M. F. (2013) · 2013 Humaine Association Conference on Affective Computing and Intelligent Interaction · doi:10.1109/acii.2013.65 · Cited by 201
He, L., Jiang, D., Yang, L., Pei, E., Wu, P., & Sahli, H. (2015) · Proceedings of the 5th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2808196.2811641 · Cited by 147 · cited 2×
“Compared to SVR, LSTM can capture the temporal information of the emotional dimension better, which is also adopted by the first place winner [23] and second place winner [24] of AVEC 2015.”
BackgroundLong Short Term Memory Recurrent Neural Network based Multimodal Dimensional Emotion Recognition
Chao, L., Tao, J., Yang, M., Li, Y., & Wen, Z. (2015) · Proceedings of the 5th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2808196.2811634 · Cited by 120 · cited 5×
“Chao et al. utilize temporal pooling function in the forward network for deep belief network [33] and LSTM [24] to reduce the label noise and redundant information.”
BackgroundMethodInfluentialSelf-citationMulti-Modal Audio, Video and Physiological Sensor Learning for Continuous Emotion Prediction
Brady, K., Gwon, Y., Khorrami, P., Godoy, E., Campbell, W., Dagli, C., & Huang, T. S. (2016) · Proceedings of the 6th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2988257.2988264 · Cited by 110 · cited 1×
“The winner of AVEC 2016 [27] extracts MFCC features as one type of audio features which achieve better performance than other audio features in arousal and valence dimension.”
BackgroundCorrecting Time-Continuous Emotional Labels by Modeling the Reaction Lag of Evaluators
Mariooryad, S., & Busso, C. (2015) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2014.2334294 · Cited by 106 · cited 2×
“Soroosh et al. [31] propose to compensate for this reaction lag by finding the time-shifting that maximizes the mutual information between the expressive behaviors and the time-continuous annotations.”
BackgroundopenXBOW - Introducing the Passau Open-Source Crossmodal Bag-of-Words ToolkitPreprint
Schmitt, M., & Schuller, B. W. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1605.06778 · Cited by 95 · cited 1×
“Finally, bag-of-text-words (BoTW) features are extracted by openXBOW [30].”
MethodMultimodal Emotion Recognition for AVEC 2016 Challenge
Povolny, F., Matejka, P., Hradis, M., Popková, A., Otrusina, L., Smrz, P., Wood, I., Robin, C., & Lamel, L. (2016) · Proceedings of the 6th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2988257.2988268 · Cited by 80 · cited 4×
“Filip et al. [17][18] adopt two Neural Networks (NNs) structure, the output of the first network is stacked in time, defining context-dependent input features for the second NN.”
BackgroundInfluentialInsights into deep neural networks for speaker recognition
Garcia-Romero, D., & McCree, A. V. (2015) · Interspeech (USB) · doi:10.21437/interspeech.2015-298 · Cited by 51 · cited 1×
“Bottle-neck (BN) features are firstly proposed to improve the performance of speech recognition [14] and then found to be effective in language identification [15] and speaker identification [16].”
BackgroundAVEC 2015: The 5th International Audio/Visual Emotion Challenge and Workshop
Ringeval, F., Schuller, B., Valstar, M., Cowie, R., & Pantic, M. (2015) · Proceedings of the 23rd ACM international conference on Multimedia · doi:10.1145/2733373.2806408 · Cited by 50 · cited 5×
“We handle origin features to obtain robust and efficient features [9].”
BackgroundMethodInfluentialMultilingual bottleneck features for language recognition
Fér, R., Matějka, P., Grézl, F., Plchot, O., & Černocký, J. (2015) · Interspeech (USB) · doi:10.21437/interspeech.2015-161 · Cited by 42 · cited 1×
“Bottle-neck (BN) features are firstly proposed to improve the performance of speech recognition [14] and then found to be effective in language identification [15] and speaker”
BackgroundMulti-scale Temporal Modeling for Dimensional Emotion Recognition in Video
Chao, L., Tao, J., Yang, M., Li, Y., & Wen, Z. (2014) · Proceedings of the 4th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2661806.2661811 · Cited by 39 · cited 1×
“Chao et al. utilize temporal pooling function in the forward network for deep belief network [33] and LSTM [24] to reduce the label noise and redundant information.”
BackgroundSelf-citationExploring Multimodal Visual Features for Continuous Affect Recognition
Sun, B., Cao, S., Li, L., He, J., & Yu, L. (2016) · Proceedings of the 6th International Workshop on Audio/Visual Emotion Challenge · doi:10.1145/2988257.2988270 · Cited by 31 · cited 3×
“As previously mentioned, Bo et al. [19] verify the effectiveness of deep visual features based on AlexNet and ResNet for continuous emotion recognition.”
BackgroundRevisiting the EmotiW challenge: how wild is it really?
Kächele, M., Schels, M., Meudt, S., Palm, G., & Schwenker, F. (2016) · Journal on Multimodal User Interfaces · doi:10.1007/s12193-015-0202-7 · Cited by 23 · cited 1×
“The past emotion recognition focused on laboratory settings, which is difficult to be applied in real world situations effectively [2].”
BackgroundGrézl, F., Egorova, E., & Karafiát, M. (2014) · IEEE Spoken Language Technology Workshop (SLT) · doi:10.1109/slt.2014.7078548 · Cited by 22 · cited 1×
“Bottle-neck (BN) features are firstly proposed to improve the performance of speech recognition [14] and then found to be effective in language identification [15] and speaker identification [16].”
BackgroundInvestigation of Bottle-Neck Features for Emotion Recognition
Popková, A., Povolný, F., Matějka, P., Glembek, O., Grézl, F., & Černocký, J. (2016) · Lecture notes in computer science · doi:10.1007/978-3-319-45510-5_49 · Cited by 7 · cited 3×
“Filip et al. [17][18] adopt two Neural Networks (NNs) structure, the output of the first network is stacked in time, defining context-dependent input features for the second NN.”
Background
Research works printed but not linked: 1
- 10.1016/j.imavis.2012.03.001 · DOI · Paper page
Cited by
Disentangled Representation Learning for Multimodal Emotion Recognition
Yang, D., Huang, S., Kuang, H., Du, Y., & Zhang, L. (2022) · Proceedings of the 30th ACM International Conference on Multimedia · doi:10.1145/3503161.3547754 · Cited by 266
“Benefiting from the excellent performance of deep learning technologies in processing diverse signals [8, 21, 26– 28, 46, 54], various models have been developed to extract emotionrelated information from multimodal sequences, such as convolution neural networks [20], recurrent neural networks [22, 41], transformers […”
BackgroundA Survey of Deep Learning-Based Multimodal Emotion Recognition: Speech, Text, and Face
Lian, H., Lu, C., Li, S., Zhao, Y., Tang, C., & Zong, Y. (2023) · Entropy · doi:10.3390/e25101440 · Cited by 224
“Furthermore, Huang et al. [86] adopted Long Short-Term Memory (LSTM) networks to create affect prediction models that fully leverage affective information from each modality.”
BackgroundMultimodal Transformer Fusion for Continuous Emotion Recognition
Huang, J., Tao, J., Liu, B., Lian, Z., & Niu, M. (2020) · ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp40776.2020.9053762 · Cited by 173
“Traditional feature level fusion directly feeds the concatenated features into a classifier or uses shallow-layered fusion models [5], but it has the difficulty to learn mutual relationships among different modalities.”
BackgroundSelf-citationMocanu, B., Țapu, R., & Zaharia, T. B. (2023) · Image and Vision Computing · doi:10.1016/j.imavis.2023.104676 · Cited by 182
Transformer Encoder With Multi-Modal Multi-Head Attention for Continuous Affect Recognition
Chen, H., Jiang, D., & Sahli, H. (2020) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2020.3037496 · Cited by 158
Metadata
- Authors
- Huang, J., Li, Y., Tao, J., Lian, Z., Wen, Z., Yang, M., & Yi, J.
- Affiliations
- Chinese Academy of Sciences (CN)
- Venue
- Proceedings of the 7th Annual Workshop on Audio/Visual Emotion Challenge
- Publisher
- ACM
- Published
- 2017-10-23
- Pages
- 11-18
- OpenAlex
- W2766925079
- Cited by
- 54
- References
- 34
- Citations by year
- 2025: 3 · 2023: 10 · 2022: 14 · 2021: 9 · 2020: 7 · 2019: 8 · 2018: 3
- FWCI
- 5.65
- Type
- conference paper
- Language
- en
- Open access
- closed
- Funders
- the National Key Research & Development Plan of China (No. 2016YFB1001404)
- the Major Program for the National Social Science Fund of China (13&ZD189)
- the National High-Tech Research and Development Program of China (863 Program) (No.2015AA016305)
- the National Natural Science Foundation of China (NSFC) (No.61425017)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Physical Sciences › Computer Science › Artificial Intelligence › Sentiment Analysis and Opinion Mining
- Keywords
- audio-visual emotion recognition, acoustic features, appearance features, LSTM-RNN, support vector regression, decision fusion, valence and arousal, liking, temporal pooling
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.