Deep Implicit Distribution Alignment Networks for cross-Corpus Speech Emotion Recognition
The DOI matches its Crossref record; 22 of the 23 references it cites were found in a registry or an index.
Abstract
In this paper, we propose a novel deep transfer learning method called deep implicit distribution alignment networks (DIDAN) to deal with cross-corpus speech emotion recognition (SER) problem, in which the labeled training (source) and unlabeled testing (target) speech signals come from different corpora. Specifically, DIDAN first adopts a simple deep regression network consisting of a set of convolutional and fully connected layers to directly regress the source speech spectrums into the emotional labels such that the proposed DIDAN can own the emotion discriminative ability. Then, such ability is transferred to be also applicable to the target speech samples regardless of corpus variance by resorting to a well-designed regularization term called implicit distribution alignment (IDA). Unlike widely-used maximum mean discrepancy (MMD) and its variants, the proposed IDA absorbs the idea of sample reconstruction to implicitly align the distribution gap, which enables DIDAN to learn both emotion discriminative and corpus invariant features from speech spectrums. To evaluate the proposed DIDAN, extensive cross-corpus SER experiments on widely-used speech emotion corpora are carried out. Experimental results show that the proposed DIDAN can outperform lots of recent state-of-the-art methods in coping with the cross-corpus SER tasks.
Verification finished with gaps
Some checks didn’t finish. The results below cover the steps that did.
- Registry recordVerified
- Full textRead the full text
- Reference list22 references found
- Reference check22 of 22 references checked
- StatisticsNeeds the full text
- ClaimsDidn’t finish — the rest of the report still applies.
- Cited sourcesDidn’t finish — the rest of the report still applies.
- Self-checkChecked
- VerificationFindings verified
Overview
CoChat Summary
TL;DR
The paper proposes Deep Implicit Distribution Alignment Networks (DIDAN), a deep transfer learning method for cross-corpus speech emotion recognition that aligns source and target feature distributions by enforcing target features to be sparsely reconstructed from source features. Across six cross-corpus tasks on EmoDB, eNTERFACE, and CASIA, DIDAN achieved the best average UAR of 39.8%.
Key findings
- DIDAN achieved the best average UAR of 39.8% among all compared methods and outperformed all comparison methods on three of six tasks: E→B (46.0%), B→C (39.1%), and C→B (54.5%).
- On the remaining three tasks, DIDAN was competitive but not best: B→E 36.1% vs. JIASL 36.9%, E→C 31.9% vs. GFK 32.8%, and C→E 30.9% vs. JIASL 33.2%.
- Ablation on B→E, E→B, and B→C showed IDA improved over VGG-11 without IDA (32.8%, 38.8%, 36.4%) and over DIDAN without sparsity regularization (34.5%, 42.9%, 37.7%), reaching 36.1%, 46.0%, and 39.1% with full IDA.
- DIDAN also outperformed DAN using MMD (35.2%, 39.2%, 36.7%) and DANN using adversarial learning (35.0%, 43.6%, 37.7%) on those three ablation tasks.
Methods
- DIDAN combines a deep regression network, consisting of convolutional and fully connected layers, with an implicit distribution alignment (IDA) regularization term that sparsely reconstructs target speech features from source speech features.
- Experiments used three public speech emotion corpora—EmoDB (German), eNTERFACE (English audio only), and CASIA (Chinese)—to construct six cross-corpus SER tasks: B→E, E→B, B→C, C→B, E→C, and C→E.
- Performance was measured by unweighted average recall (UAR); deep methods used VGG-11 as backbone with speech spectrums resized to 224×224, while subspace learning methods used IS09 features from openSMILE.
Limitations
- The paper does not state an explicit limitations section.
- The text states that nearly all methods, including DIDAN, cannot well cope with cross-corpus SER tasks between eNTERFACE and CASIA, attributed to large differences between the English induced-paradigm corpus and the Chinese simulated-emotion corpus.
- Evident from the reported results: DIDAN did not achieve the best UAR on B→E, E→C, or C→E, and the evaluation is limited to three corpora and six cross-corpus tasks.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
This check didn’t finish.
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Currently, a common approach involves using self-supervised training methods to learn high-quality representations from unlabeled data, and then using these representations to replace hand-crafted features to complete downstream tasks. This is a research hotspot in the field of speech emotion [49,50,51,52,53].”
From A Survey of Deep Learning-Based Multimodal Emotion Recognition: Speech, Text, and Face (2023) · cites this paper as [52]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.3. Results and Discussions
“| | DANN | 35.0 | 43.6 | 37.6 | 52.3 | 28.9 | 30.0 | 37.9 |”
§ 3.3. Results and Discussions
“| DANN (Adversarial Learning) | 35.0 | 43.6 | 37.7 |”
Citations
Cites
Domain Adaptation via Transfer Component Analysis
Pan, S. J., Tsang, I. W., Kwok, J. T., & Yang, Q. (2011) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.2010.2091281 · Cited by 4,753 · cited 3×
“Specifically, for TCA, GFK, and SA, we search its reduced feature dimension from [5 : 5 : d max ] .”
Learning Transferable Features with Deep Adaptation Networks
Long, M., Cao, Y., Wang, J., & Jordan, M. I. (2015) · arXiv · doi:10.48550/arxiv.1502.02791 · Cited by 4,505 · cited 1×
“As for the deep learning ones, deep adaptation network (DAN) [21], domain-adversarial neural network (DANN) [22], and conditional domain adversarial network (CDAN) [23], and deep subdo-main adaptation network (DSAN) [24] are adopted.”
MethodSelf-citationOpensmile: 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×
“Note that for these subspace learning methods, we use openSMILE toolkit [19] to extract IS09 speech feature sets [20], which consists of 32 low-level acoustic descriptors and 12 statistical functions, to describe the speech signals in all three speech corpora.”
MethodA database of German emotional speech
Burkhardt, F., Paeschke, A., Rolfes, M., Sendlmeier, W. F., & Weiss, B. (2005) · Interspeech 2005 · doi:10.21437/interspeech.2005-446 · Cited by 2,365 · cited 1×
“, EmoDB (B) [11], eNTERFACE (E) [12], and CASIA (C) [13], are employed to design the cross-corpus SER experiments.”
MethodGeodesic flow kernel for unsupervised domain adaptation
Gong, B., Shi, Y., Sha, F., & Grauman, K. (2012) · 2012 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2012.6247911 · Cited by 2,093 · cited 5×
“%) in B → E and GFK (32.8”
MethodInfluentialSurvey on speech emotion recognition: Features, classification schemes, and databases
El Ayadi, M., Kamel, M. S., & Karray, F. (2011) · Pattern Recognition · doi:10.1016/j.patcog.2010.09.020 · Cited by 2,073
Unsupervised Visual Domain Adaptation Using Subspace Alignment
Fernando, B., Habrard, A., Sebban, M., & Tuytelaars, T. (2013) · 2013 IEEE International Conference on Computer Vision · doi:10.1109/iccv.2013.368 · Cited by 1,314 · cited 1×
“The transfer subspace learning methods include transfer component analysis (TCA) [14], geodesic flow kernel (GFK) [15], subspace alignment (SA) [16], domainadaptive subspace learning (DoSL) [7], joint distribution adaptive regression (JDAR) [17] and joint distribution implicitly aligned subspace learning (JIASL) [18].”
MethodConditional Adversarial Domain Adaptation
Long, M., Cao, Z., Wang, J., & Jordan, M. I. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1705.10667 · Cited by 1,281
Self-citationDeep Subdomain Adaptation Network for Image Classification
Zhu, Y., Zhuang, F., Wang, J., Ke, G., Chen, J., Bian, J., Xiong, H., & He, Q. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.2988928 · Cited by 1,245 · cited 2×
“As for the deep learning ones, deep adaptation network (DAN) [21], domain-adversarial neural network (DANN) [22], and conditional domain adversarial network (CDAN) [23], and deep subdo-main adaptation network (DSAN) [24] are adopted.”
Self-citationThe INTERSPEECH 2009 emotion challenge
Schuller, B., Steidl, S., & Batliner, A. (2009) · Interspeech 2009 · doi:10.21437/interspeech.2009-103 · Cited by 938 · cited 1×
“Note that for these subspace learning methods, we use openSMILE toolkit [19] to extract IS09 speech feature sets [20], which consists of 32 low-level acoustic descriptors and 12 statistical functions, to describe the speech signals in all three speech corpora.”
Method
Research works printed but not linked: 2
- tao, 2008, Design of speech corpus for mandarin text to speech, The Blizzard Challenge 2008 workshop, 1
- martin, 2006, The enterface’05 audio-visual emotion database, ICDE Workshops, 8
Cited by
A 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
Self-citationSönmez, Y. Ü., & Varol, A. (2024) · Intelligent Systems with Applications · doi:10.1016/j.iswa.2024.200351 · Cited by 65
Improving Speaker-Independent Speech Emotion Recognition using Dynamic Joint Distribution Adaptation
Lu, C., Zong, Y., Lian, H., Zhao, Y. D., Schuller, B. W., & Zheng, W. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10447452 · Cited by 15
“Fortunately, recent SER studies have demonstrated that Domain Adaptation (DA) holds promise in mitigating the domain/speaker bias in SER [7], [8], [9].”
BackgroundSelf-citationLin, H.-C., Lin, Y.-C., Chou, H.-C., & Lee, H.-Y. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10887615 · Cited by 24
“Domain adaptation methods focus on aligning features across languages through techniques such as feature normalization [19], [34], direct alignment [18], [35], [43], and domain adversarial learning [43].”
Emotion-Aware Contrastive Adaptation Network for Source-Free Cross-Corpus Speech Emotion Recognition
Zhao, Y. D., Wang, J., Lu, C., Li, S., Schuller, B. W., Zong, Y., & Zheng, W. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10446541 · Cited by 8
“Notably, even without the presence of source data, ECAN achieves comparable performance to source-available meth-ods such as DIDAN [6] and DTTRN [7].”
BackgroundMethodSelf-citation
Metadata
- Authors
- Zhao, Y., Wang, J., Zong, Y., Zheng, W., Lian, H., & Zhao, L.
- Affiliations
- Southeast University (CN)
- Venue
- ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
- Publisher
- IEEE
- Published
- 2023-06-04
- Pages
- 1-5
- ISSN
- 1520-6149 (ISSN-L)
- OpenAlex
- W4372263897
- Cited by
- 19
- References
- 23
- Citations by year
- 2026: 3 · 2025: 5 · 2024: 8 · 2023: 3
- FWCI
- 7.01
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE International Conference on Acoustics Speech and Signal Processing (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Keywords
- domain adaptation, transfer learning, sample reconstruction, maximum mean discrepancy
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Deep Implicit Distribution Alignment Networks for cross-Corpus Speech Emotion Recognition. https://app.cochat.ai/papers/10.1109/icassp49357.2023.10095388
- The paper
- Zhao, Y., Wang, J., Zong, Y., Zheng, W., Lian, H., & Zhao, L. (2023). Deep Implicit Distribution Alignment Networks for cross-Corpus Speech Emotion Recognition. ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). https://doi.org/10.1109/icassp49357.2023.10095388
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.