Pith. sign in

REVIEW 4 major objections 8 minor 47 references

Controlling for Confounders in Multimodal Emotion Classification via Adversarial Learning

T0 review · 4 major / 8 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Emotion recognition models trained to unlearn stress generalize better to new domains than models that learn stress cues.

desk verdict Useful empirical study of adversarial stress decorrelation for emotion recognition, but the MuSE stress labels are too confounded with session/time to pin the gains on 'stress' specifically. read the letter →

arxiv 1908.08979 v1 pith:OVLN24M5 submitted 2019-08-23 cs.LG cs.CLcs.SDeess.ASstat.ML

Mimansa Jaiswal , Zakaria Aldeneh , Emily Mower Provost This is my paper · ORCID
classification cs.LGcs.CLcs.SDeess.ASstat.ML
keywords adversariallearningemotionrecognitionstressconfoundersdomaingeneralizationgradientreversalmultimodalaffectivecomputing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that stress acts as a confounder in emotion recognition: models trained to read emotion from voice and text silently learn stress-related cues, and those cues do not transfer well to new settings. The authors show that stress is detectable inside emotion-trained representations, and that forcing a stress classifier to chance level during training, by reversing its gradient, makes the emotion model generalize better to unseen datasets than the same model trained without that step. This matters because emotion recognition systems built on controlled laboratory data are likely to fail when deployed with speakers who are stressed, tired, or otherwise psychologically different. The claim is that explicitly unlearning such extraneous factors is a viable path toward more dependable affective computing.

What carries the argument

The central mechanism is the Gradient Reversal Layer (GRL), placed between the shared embedding sub-network and the adversarial stress classifier. During the forward pass the GRL acts as identity, but during backpropagation it multiplies the stress-classifier gradients by a negative constant, $\lambda$, so the embedding learns to keep emotion information while making stress unclassifiable. This same machinery is reused with a spontaneity classifier on IEMOCAP to show that the approach extends beyond stress.

What would settle it

Train the same adversarial architecture with random session-level labels, or with session identity as the adversarial target, and compare cross-dataset UAR; if the gains match those obtained with true stress labels, the claim that stress specifically is the confounder is not supported. Alternatively, collect per-utterance stress ratings on MuSE and rerun the MuSE-to-IEMOCAP transfer to see whether the improvements survive with finer-grained stress labels.

Watch

Extended reading notes

Core claim

The paper's central claim is that emotion recognition models become more generalizable when stress is actively removed from the learned emotion representation during training. Using a gradient reversal layer, the network is trained to predict emotion while an adversarial stress classifier is forced to perform at chance, decorrelating stress information from the acoustic and lexical features that carry emotion. The results show that stress is indeed encoded in emotion-trained representations, more so in acoustic features for activation and in lexical features for valence. Controlling for stress lowers within-corpus performance slightly, but consistently improves cross-dataset performance: for example, MuSE-to-IEMOCAP activation UAR rises from 0.419 to 0.448 for acoustic input, and valence from 0.431 to 0.472 for multimodal input. The paper concludes that extraneous psychological factors such as stress should be explicitly accounted for when building and testing emotion recognition models.

Load-bearing premise

The paper's results hinge on treating MuSE's session-level stress scores as valid per-utterance stress ground truth; because stressed sessions were recorded during final exams and non-stressed sessions after exams, the adversarial head may be removing time or session cues rather than stress itself.

Editorial extensions

If this is right

  • Emotion recognition systems that control for stress during training show significant cross-corpus gains, for instance MuSE-to-IEMOCAP activation improving from 0.419 to 0.448 UAR for acoustic input and valence from 0.431 to 0.472 for multimodal input.
  • Stress decorrelation costs some source-domain performance, but the transfer gains indicate the removed signal was partly a dataset-specific shortcut rather than essential emotion information.
  • High stress levels harm emotion classification more than low stress, with the largest drop for mid activation (22.03% relative accuracy loss) and low valence (8.22%), suggesting that stress-aware training matters most in high-stress conditions.
  • The same adversarial recipe transfers to other confounders: decorrelating spontaneity in IEMOCAP improves cross-dataset performance on MuSE and MSP-Improv for several setups.
  • Lexical markers such as fillers, adverbs, and content rate positively correlate with samples that benefit from stress decorrelation, providing interpretable signals for when the adversarial model is more trustworthy.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because MuSE stress labels are assigned per session and stressed sessions were recorded during final exams, the adversarial head may actually be unlearning session identity or time-of-semester cues rather than stress per se; training the same architecture against random session labels would test whether stress is the active ingredient.
  • If the session-label confound is real, the architecture becomes a general-purpose nuisance-factor remover: it could decorrelate recording room, day, speaker cohort, or any coarse-grained grouping, with consequences beyond emotion recognition.
  • The adjusted-probability-of-success analysis suggests a deployable per-sample reliability triage: on new data, trust the adversarially trained model more for utterances rich in fillers and adverbs, and the normal model elsewhere, a rule that could be validated in a prospective study.
  • The method likely extends to other psychological states such as anxiety, fatigue, or trust, but those extensions would need per-utterance labels or carefully controlled elicitation to avoid the same session-level critique.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 8 minor

Summary. The paper studies whether stress acts as a confounder in multimodal emotion recognition and proposes an adversarial gradient-reversal training scheme that removes stress information from learned acoustic and lexical representations. Using three datasets (MuSE, IEMOCAP, MSP-Improv), the authors report that emotion-trained representations encode stress, that adversarial stress decorrelation reduces within-domain emotion accuracy, and that the adversarially trained models often improve cross-dataset generalization (MuSE to IEMOCAP/MSP-Improv, and IEMOCAP to MuSE/MSP-Improv for the spontaneity confounder). They also analyze lexical patterns (LIWC categories, fillers, content rate) that correlate with improvements under adversarial training. The central claim is that controlling for stress during training yields emotion recognition models that generalize better to new domains than models that do not control for stress.

Significance. If the central claim holds, the work is significant for affective computing: it identifies a real extraneous psychological factor and offers a concrete, architecture-level remedy with evidence across multiple corpora. Strengths of the paper include its cross-dataset evaluation, the use of speaker-independent partitions, multiple random seeds with averaged predictions, and Benjamini-Hochberg-corrected correlation analyses. The method is described sufficiently to be reimplemented. The main weakness is that the MuSE stress labels are session-level and confounded with recording period and speaker identity, so the stress-specific interpretation of the observed generalization gains is not yet established; this directly affects the primary claim in the abstract and conclusions.

major comments (4)
  1. [Section 3 (MuSE) and Section 4 (Stress-Invariance), Table 4] The stress labels in MuSE are session-level: each participant contributes one stressed session (recorded during final exams) and one non-stressed session (recorded after exams), and the same stress label is assigned to every utterance from a session. The adversarial GRL therefore minimizes the predictability of session identity and exam period, not necessarily the psychological construct of stress per se. Because the cross-dataset generalization results in Table 4 are trained on MuSE and tested on IEMOCAP/MSP-Improv, the observed UAR improvements could be due to removal of session- or time-specific acoustic and lexical covariates rather than to removal of stress. To support the stress-specific conclusion, the authors should compare against a control that removes session or speaker identity (e.g., an adversarial speaker classifier), or otherwise demonstrate that the gains are not obtained when the adversarial target is replaced by a session-related nuisance label.
  2. [Table 3] The third row of Table 3 repeats the first row exactly ('Train: Stress (Medium + High) Test: Stress (Low)'), and no row reports the high-stress target condition. The text in Question 4 states 'Considering high levels of stress as our target, adversarial classification significantly improves performance over normal classification for acoustic setups for activation and for all setups for valence,' but this claim is not supported by the table as printed. The missing high-stress partition should be reported, or the text corrected to match the available rows.
  3. [Section 4 (Training) and Tables 4 and 5] The statistical test details for the cross-dataset results are missing. The tables use bold to mark statistically significant differences, and the text mentions paired t-tests, but it is not clear what the paired observations are: utterances, speakers, random-seed runs, or bootstrap resamples. Since the cross-dataset experiments train on one full source corpus and test on one target corpus, a paired t-test over utterances would ignore speaker/session clustering, while a paired test over the three random seeds would have n=3. Confidence intervals are not reported. The authors should specify the exact test procedure, the units of analysis, and the number of paired observations for Tables 4 and 5.
  4. [Section 4 (Training recipe) and Table 1] The model selection procedure chooses hyperparameters (including the GRL weight lambda) so that the validation stress UAR is approximately at chance (0.33), and the resulting stress UAR values in Table 1 are then presented as evidence that stress has been 'unlearned.' This is partially circular: the selection criterion directly enforces the reported stress classification outcome. The emotion classification results are still interpretable, but the stress-decorrelation evidence would be stronger if the stress UAR were reported on a held-out test set not used for model selection, or if the paper explicitly argued why the validation-based selection does not compromise the comparison.
minor comments (8)
  1. [Abstract] The sentence 'We conclude that is is necessary' contains a typo; it should read 'it is necessary.'
  2. [Section 3 (Labels, Stress Labels)] The sentence reporting the paired t-test results ('the scores are significantly different for both sets (16.11 vs 18.53)') does not clarify what the numbers represent; please state that they are mean PSS scores for the non-stressed and stressed sessions, and report the t-statistic and p-value.
  3. [Section 3 (Labels, Emotion Labels)] The MuSE emotion binning uses boundaries on a nine-point Likert scale (low: [min,4.5], mid: (4.5,5.5], high: (5.5,max]), while IEMOCAP and MSP-Improv bin a five-point scale with different boundaries. This means the class distributions and label semantics differ across corpora; please make explicit that cross-dataset results are with respect to each corpus's own binning and discuss any consequences for interpreting transfer performance.
  4. [Table 3] In addition to the duplicated third row, the partition labels are ambiguous because the training set is always described as two stress levels and the test set as one; adding the missing high-stress row and clearly labeling all three partitions would improve readability.
  5. [Section 5, Question 4] The phrase 'We train a dataset on complete MuSE data' should be 'We train a model on complete MuSE data.'
  6. [Section 5, Question 6] The definition of adjusted probability of success contains a typo 'Pnor ma,sl(Success)'; this should be 'P_normal,s(Success).'
  7. [Section 5, Question 6] There are typos in the headings and text: 'spontatenity' should be 'spontaneity,' 'classifaction' should be 'classification,' and 'certainity' should be 'certainty.'
  8. [References] References [30] and [31] refer to the same McHardy, Adel, and Klinger paper (the arXiv preprint and the CoRR version); please consolidate them into a single citation.

Circularity Check

1 steps flagged · score 2.0 of 10

One stress-decorrelation result is partly a selection target, but the cross-corpus generalization claim is independently tested.

  1. fitted input called prediction [Section 4 (Training) and Section 5 (Question 2), Tables 1a/1b]
    "For the adversarial classification model, we ensure that the chosen model yields a validation unweighted average recall (UAR) that is random (0.33) for the stress classification task. For the adversarial emotion classification setups, we use the hyper-parameters that maximize the validation emotion classification performance while minimizing the validation stress classification performance."

    The tables' adversarial stress UAR rows (roughly 0.305-0.332) are presented as evidence that stress has been decorrelated from the emotion representations. But those numbers are not a free prediction: the training recipe explicitly selects hyperparameters so that validation stress UAR is at chance (0.33) and minimizes validation stress performance. Thus the near-chance stress accuracy is an artifact of the selection criterion, not an independent confirmation of decorrelation. This makes the stress-decorrelation part of the analysis partially circular. The paper's main generalizability claim is nevertheless independent: Table 4's IEMOCAP and MSP-Improv results are external benchmarks not used in the selection, so they do not reduce to the stress-chance objective.

full rationale

The central claim—that adversarial stress removal improves emotion recognition transfer—is evaluated on held-out external corpora (IEMOCAP, MSP-Improv) and on stress-partitioned MuSE folds, so it does not reduce to the fitting procedure. The only circularity I can exhibit is the use of the chance-level validation stress UAR as a model-selection criterion and then as evidence that stress is decorrelated; the resulting near-random stress accuracies in Tables 1a/1b restate that selection target. This does not infect the external-transfer results, and the paper cites standard GRL methodology rather than importing a uniqueness claim from the authors. Overall score 2.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a small number of data-construction choices and domain assumptions. The stress labels are constructed with hand-chosen cutoffs and an ad hoc question weighting, and they are assigned at session level. The adversarial method assumes gradient reversal can remove the labeled confounder without removing emotion information. No new entities or physical objects are introduced.

free parameters (6)
  • Stress binning thresholds (mean - 2 and mean + 2) = Not reported; defined relative to the mean of adjusted PSS scores
    The three stress classes are defined by cutoffs at mean - 2 and mean + 2. The choice of 2 is not justified by theory and directly determines the labels used to train the adversarial stress classifier.
  • PSS question 3 weight multiplier = 2
    The authors double the weight of question 3 because it showed a significant difference between stressed and non-stressed conditions, an ad hoc adjustment that changes the stress scores and thus the stress labels.
  • Emotion binning thresholds for MuSE = low: [min, 4.5]; mid: (4.5, 5.5]; high: (5.5, max]
    Continuous averaged annotations are binned into three classes with hand-chosen cutoffs, which sets the target labels for emotion classification.
  • Emotion binning thresholds for IEMOCAP and MSP-Improv = low: [1, 2.75]; mid: (2.75, 3.25]; high: (3.25, max]
    Same binning procedure as MuSE, applied to the five-point scale.
  • GRL gradient reversal weight lambda = Selected from {0.3, 0.6, 0.8} per setup
    Lambda is chosen on validation to maximize emotion UAR while forcing stress validation UAR to chance. This controls the strength of stress removal and is tuned to the data.
  • Network architecture hyperparameters (conv layers, kernels, widths, GRU layers, dense layers) = Not reported; selected from search ranges via validation
    The search ranges are given in Section 4, Training, but the final selected values for each setup are not reported, which affects reproducibility.
assumptions (5)
  • domain assumption Self-reported Perceived Stress Scale scores are a valid measure of a speaker's stress state at the time of recording, and the session-level label applies to every utterance in that session.
    Section 3 introduces PSS scores and assigns the same stress label to all utterances from the same session. The adversarial training and the 'stress is encoded' analyses depend on these labels being accurate per utterance.
  • domain assumption Gradient reversal with a properly chosen lambda removes confounder information from the learned representation while preserving emotion-discriminative information.
    The method rests on the premise that a stress classifier at chance on validation indicates the representation is stress-invariant; see Section 4, Stress-Invariance.
  • domain assumption Stress is a confounder that influences both the input features (acoustic and lexical) and the emotion labels or their perception.
    The motivation in Section 1 and the psychological citations support this, but it is not directly tested in this work.
  • domain assumption Averaged crowd-sourced emotion annotations, binned into low/mid/high, are an appropriate ground truth for emotion.
    Section 3, Labels; all emotion classification experiments rely on this.
  • standard math Unweighted average recall is an appropriate metric for the class-imbalanced emotion and stress classification tasks.
    Section 4 uses UAR following [39]; it is a standard choice but affects how results are compared.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Controlling for Confounders in Multimodal Emotion Classification via Adversarial Learning." pith.science (2026). https://pith.science/paper/OVLN24M5

@misc{pith2026190808979,
  author       = {Pith},
  title        = {Pith review of: Controlling for Confounders in Multimodal Emotion Classification via Adversarial Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OVLN24M5}},
  note         = {Machine review of arXiv:1908.08979}
}
read the original abstract

Various psychological factors affect how individuals express emotions. Yet, when we collect data intended for use in building emotion recognition systems, we often try to do so by creating paradigms that are designed just with a focus on eliciting emotional behavior. Algorithms trained with these types of data are unlikely to function outside of controlled environments because our emotions naturally change as a function of these other factors. In this work, we study how the multimodal expressions of emotion change when an individual is under varying levels of stress. We hypothesize that stress produces modulations that can hide the true underlying emotions of individuals and that we can make emotion recognition algorithms more generalizable by controlling for variations in stress. To this end, we use adversarial networks to decorrelate stress modulations from emotion representations. We study how stress alters acoustic and lexical emotional predictions, paying special attention to how modulations due to stress affect the transferability of learned emotion recognition models across domains. Our results show that stress is indeed encoded in trained emotion classifiers and that this encoding varies across levels of emotions and across the lexical and acoustic modalities. Our results also show that emotion recognition models that control for stress during training have better generalizability when applied to new domains, compared to models that do not control for stress during training. We conclude that is is necessary to consider the effect of extraneous psychological factors when building and testing emotion recognition models.

Figures

Figures reproduced from arXiv: 1908.08979 by the authors.

Figure 1
Figure 1. Adversarial multi-task network architecture. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 40 canonical work pages

  1. [1]

    Mohammed Abdelwahab and Carlos Busso. 2018. Domain adversarial for acoustic emotion recognition. IEEE/ACM Transactions on Audio, Speech, and Language Processing 26, 12 (2018), 2423–2435

    work page 2018

  2. [2]

    Arthur Aron, Edward Melinat, Elaine N Aron, Robert Darrin Vallone, and Renee J Bator. 1997. The experimental generation of interpersonal closeness: A procedure and some preliminary findings. Personality and Social Psychology Bulletin 23, 4 (1997), 363–377

    work page 1997

  3. [3]

    Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman Vaughan. 2010. A theory of learning from different domains. Machine learning 79, 1-2 (2010), 151–175

    work page 2010

  4. [4]

    Yoav Benjamini and Yosef Hochberg. 1995. Controlling the false dis- covery rate: a practical and powerful approach to multiple testing. Journal of the Royal statistical society: series B (Methodological) 57, 1 (1995), 289–300

    work page 1995

  5. [5]

    Tony W Buchanan, Jacqueline S Laures-Gore, and Melissa C Duff. 2014. Acute stress reduces speech fluency. Biological psychology 97 (2014), 60–66

    work page 2014

  6. [6]

    Carlos Busso, Murtaza Bulut, Chi-Chun Lee, Abe Kazemzadeh, Emily Mower, Samuel Kim, Jeannette N Chang, Sungbok Lee, and Shrikanth S Narayanan. 2008. IEMOCAP: Interactive emotional dyadic motion capture database. Language resources and evaluation 42, 4 (2008), 335

    work page 2008

  7. [7]

    Carlos Busso, Srinivas Parthasarathy, Alec Burmania, Mohammed AbdelWahab, Najmeh Sadoughi, and Emily Mower Provost. 2017. MSP- IMPROV: An acted corpus of dyadic interactions to study emotion perception. IEEE Transactions on Affective Computing 8, 1 (2017), 9 67–80

    work page 2017

  8. [8]

    FranÃğois Chollet. 2015. keras. https://github.com/fchollet/keras

    work page 2015

Show all 47 references
  1. [9]

    Herbert H Clark and Jean E Fox Tree. 2002. Using uh and um in spontaneous speaking. Cognition 84, 1 (2002), 73–111

    2002

  2. [10]

    Sheldon Cohen, Tom Kamarck, and Robin Mermelstein. 1983. A global measure of perceived stress. Journal of health and social behavior (1983), 385–396

    1983

  3. [11]

    Daniel J DeDora, Joshua M Carlson, and Lilianne R Mujica-Parodi

  4. [12]

    Duvall, Alan Stuart Robbins, Thomas R Graham, and Scott Divett

    Emily D. Duvall, Alan Stuart Robbins, Thomas R Graham, and Scott Divett. 2014. Exploring Filler Words and Their Impact

    2014

  5. [13]

    Yanai Elazar and Yoav Goldberg. 2018. Adversarial Removal of Demo- graphic Attributes from Text Data. arXiv preprint arXiv:1808.06640 (2018)

    2018 arXiv

  6. [14]

    Thomas W Freeman, John Hart, Tim Kimbrell, and Elliott D Ross

  7. [15]

    Yaroslav Ganin and Victor Lempitsky. 2014. Unsupervised do- main adaptation by backpropagation. arXiv preprint arXiv:1409.7495 (2014)

    2014 arXiv

  8. [16]

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, François Laviolette, Mario Marchand, and Victor Lempitsky. 2016. Domain-adversarial training of neural networks. The Journal of Machine Learning Research 17, 1 (2016), 2096–2030

    2016

  9. [17]

    John Gideon, Melvin G McInnis, and Emily Mower Provost. 2019. Barking up the Right Tree: Improving Cross-Corpus Speech Emotion Recognition with Adversarial Discriminative Domain Generalization (ADDoG). arXiv preprint arXiv:1903.12094 (2019)

    2019 arXiv

  10. [18]

    Guosheng Hu, Xiaojiang Peng, Yongxin Yang, Timothy M Hospedales, and Jakob Verbeek. 2017. Frankenstein: Learning deep face represen- tations using small data. IEEE Transactions on Image Processing 27, 1 (2017), 293–303

    2017

  11. [19]

    Mimansa Jaiswal, Zakaria Aldeneh, Cristian-Paul Bara, Yuanhang Luo, Mihai Burzo, Rada Mihalcea, and Emily Mower Provost. 2019. MuSE-ing on the impact of utterance ordering on crowdsourced emo- tion annotations. In 2019 IEEE International Conference on Acoustics, Speech and Sig...

    2019

  12. [20]

    Rebecca Jürgens, Annika Grass, Matthis Drolet, and Julia Fischer. 2015. Effect of acting experience on emotion expression and recognition in voice: Non-actors provide better stimuli than expected. Journal of nonverbal behavior 39, 3 (2015), 195–214

    2015

  13. [21]

    Jeffrey H Kahn, Renee M Tobin, Audra E Massey, and Jennifer A An- derson. 2007. Measuring emotional expression with the Linguistic Inquiry and Word Count. The American journal of psychology (2007), 263–286

    2007

  14. [22]

    Soheil Khorram, Zakaria Aldeneh, Dimitrios Dimitriadis, Melvin McIn- nis, and Emily Mower Provost. 2017. Capturing long-term temporal dependencies with convolutional networks for continuous emotion recognition. arXiv preprint arXiv:1708.07050 (2017)

    2017 arXiv

  15. [23]

    Soheil Khorram, Mimansa Jaiswal, John Gideon, Melvin McInnis, and Emily Mower Provost. 2018. The PRIORI Emotion Dataset: Linking Mood to Emotion Detected In-the-Wild. arXiv preprint arXiv:1806.10658 (2018)

    2018 arXiv

  16. [24]

    Yoon Kim. 2014. Convolutional neural networks for sentence classifi- cation. arXiv preprint arXiv:1408.5882 (2014)

    2014 arXiv

  17. [25]

    2008.Automatic Segmentation of Speech into Sentence-like Units

    Jáchym Kolář. 2008.Automatic Segmentation of Speech into Sentence-like Units. Ph.D. Dissertation. University of West Bohemia in Pilsen

    2008

  18. [26]

    Kalpesh Krishna, Liang Lu, Kevin Gimpel, and Karen Livescu. 2018. A Study of All-Convolutional Encoders for Connectionist Temporal Classification. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 5814–5818

    2018

  19. [27]

    Virgile Landeiro and Aron Culotta. 2016. Robust text classification in the presence of confounding bias. In Thirtieth AAAI Conference on Artificial Intelligence

    2016

  20. [28]

    Francesco Locatello, Stefan Bauer, Mario Lucic, Sylvain Gelly, Bernhard Schölkopf, and Olivier Bachem. 2018. Challenging common assump- tions in the unsupervised learning of disentangled representations. arXiv preprint arXiv:1811.12359 (2018)

    2018 arXiv

  21. [29]

    Karttikeya Mangalam and Tanaya Guha. 2017. Learning Spontane- ity to Improve Emotion Recognition in Speech. arXiv preprint arXiv:1712.04753 (2017)

    2017 arXiv

  22. [31]

    Robert McHardy, Heike Adel, and Roman Klinger. 2019. Adversarial Training for Satire Detection: Controlling for Confounding Variables. CoRR abs/1902.11145 (2019). arXiv:1902.11145 http://arxiv.org/abs/ 1902.11145

    2019 arXiv

  23. [32]

    Matthias R Mehl, Charles L Raison, Thaddeus WW Pace, Jesusa MG Arevalo, and Steve W Cole. 2017. Natural language indicators of differential gene regulation in the human immune system.Proceedings of the National Academy of Sciences 114, 47 (2017), 12554–12559

    2017

  24. [33]

    Zhong Meng, Jinyu Li, Zhuo Chen, Yang Zhao, Vadim Mazalov, Yi- fan Gang, and Biing-Hwang Juang. 2018. Speaker-invariant training via adversarial learning. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 5969–5973

    2018

  25. [34]

    Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. In Advances in neural information processing systems. 3111–3119

    2013

  26. [35]

    Pavlo Molchanov, Stephen Tyree, Tero Karras, Timo Aila, and Jan Kautz. 2016. Pruning convolutional neural networks for resource efficient transfer learning. arXiv preprint arXiv:1611.06440 3 (2016)

    2016 arXiv

  27. [36]

    John Moore, Joel Pfeiffer, Kai Wei, Rishabh Iyer, Denis Charles, Ran Gilad-Bachrach, Levi Boyles, and Eren Manavoglu. 2018. Modeling and Simultaneously Removing Bias via Adversarial Neural Networks. arXiv preprint arXiv:1804.06909 (2018)

    2018 arXiv

  28. [37]

    Silke Paulmann, Desire Furnes, Anne Ming Bøkenes, and Philip J Cozzolino. 2016. How psychological stress affects emotional prosody. Plos one 11, 11 (2016), e0165022

    2016

  29. [38]

    James W Pennebaker, Martha E Francis, and Roger J Booth. 2001. Linguistic inquiry and word count: LIWC 2001. Mahway: Lawrence Erlbaum Associates 71, 2001 (2001), 2001

    2001

  30. [39]

    Andrew Rosenberg. 2012. Classifying skewed data: Importance weight- ing to optimize average recall. In Thirteenth Annual Conference of the International Speech Communication Association

    2012

  31. [40]

    Yusuke Shinohara. 2016. Adversarial Multi-Task Learning of Deep Neural Networks for Robust Speech Recognition.. In INTERSPEECH. San Francisco, CA, USA, 2369–2372

    2016

  32. [41]

    Herman JM Steeneken and John HL Hansen. 1999. Speech un- der stress conditions: overview of the effect on speech production and on system performance. In 1999 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings. ICASSP99 (Cat. No. 99CH36258),...

    1999

  33. [42]

    Adarsh Subbaswamy, Peter Schulam, and Suchi Saria. 2018. Learning predictive models that transport. arXiv preprint arXiv:1812.04597 (2018)

    2018 arXiv

  34. [43]

    Tieleman and G

    T. Tieleman and G. Hinton. 2012. Lecture 6.5—RmsProp: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural Networks for Machine Learning. 10

    2012

  35. [44]

    Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. 2017. Adversarial discriminative domain adaptation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 7167– 7176

    2017

  36. [45]

    Wei Wang, Ivan Hernandez, Daniel A Newman, Jibo He, and Jiang Bian. 2016. Twitter analysis: Studying US weekly trends in work stress and emotion. Applied Psychology 65, 2 (2016), 355–378

    2016

  37. [46]

    Brian Hu Zhang, Blake Lemoine, and Margaret Mitchell. 2018. Mit- igating unwanted biases with adversarial learning. In Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society. ACM, 335–340. 11

    2018

  38. [2009]

    The Journal of neuropsychiatry and clinical neurosciences 21, 1 (2009), 52–58

    Comprehension of affective prosody in veterans with chronic posttraumatic stress disorder. The Journal of neuropsychiatry and clinical neurosciences 21, 1 (2009), 52–58

    2009

  39. [2011]

    Evolutionary Psychology 9, 4 (2011), 147470491100900406

    Acute stress eliminates female advantage in detection of ambiguous negative affect. Evolutionary Psychology 9, 4 (2011), 147470491100900406

    2011

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.