Emotionless: Privacy-Preserving Speech Analysis for Voice Assistants

David Boyle; Hamed Haddadi; Ranya Aloufi

REVIEW 3 major objections 6 minor 1 cited by

Emotionless: Privacy-Preserving Speech Analysis for Voice Assistants

T0 review · 3 major / 6 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 A privacy layer between the user and the cloud can convert emotional speech into neutral speech, cutting emotion-recognition accuracy by about 96% while leaving speech content and speaker identity largely intact.

desk verdict A plausible early idea for emotion-stripping voice conversion, undermined by the missing neutral-to-neutral control: the 96% drop may just be conversion artifacts, and the reported 35% WER suggests those artifacts are real. read the letter →

arxiv 1908.03632 v1 pith:VRCTY2SY submitted 2019-08-09 cs.CR cs.LGcs.SDeess.ASstat.ML

Ranya Aloufi , Hamed Haddadi , David Boyle This is my paper · ORCID

classification cs.CRcs.LGcs.SDeess.ASstat.ML

keywords speechprivacyemotionrecognitionvoiceconversionCycleGANanonymisationassistantsparalinguisticsinternetofthings

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 tries to establish that a user's emotional state can be stripped out of voice input before it is sent to a cloud assistant, while keeping enough of the speech content and the speaker's identity for the service to work. The proposed privacy layer converts emotional utterances into neutral-sounding speech with a CycleGAN-based voice-conversion network, then resynthesizes the audio with a vocoder. On a standard emotional-speech dataset, the conversion reduces emotion-recognition accuracy by about 96%, at the cost of an average word error rate of 35% for speech recognition and a roughly 0.12% degradation in speaker-recognition error rate. If true, the method offers a practical middle path between sending raw voice to the cloud and encrypting it: make the signal useless for one sensitive inference task without breaking the tasks users actually want.

What carries the argument

The machinery is a cycle-consistent generative adversarial network (CycleGAN) used as a non-parallel voice converter: two generators translate between emotional and neutral spectral-envelope features, and two discriminators plus a cycle-consistency loss force the round-trip translation to preserve content. A vocoder called WORLD supplies the acoustic front end and back end, estimating F0, spectral envelope, and aperiodicity from the raw waveform and resynthesizing speech from the converted features. This combination lets the system alter prosody-related parameters that carry emotion without requiring paired emotional and neutral recordings of the same utterance.

What would settle it

Run neutral recordings from the same dataset through the identical CycleGAN-to-WORLD conversion and score them with the same emotion classifier: if accuracy falls as much as it does for emotional recordings, the drop is a conversion artifact rather than emotion sanitization; if accuracy stays high while emotional accuracy falls, the mechanism is confirmed.

Watch

Extended reading notes

Core claim

The central claim is that sensitive paralinguistic information, specifically emotional state, can be projected out of the speech signal by mapping the acoustic parameters of emotional utterances onto those of neutral speech using an unpaired voice-conversion model. The conversion happens on the acoustic feature level: fundamental frequency, spectral envelope, and aperiodicity are extracted, the spectral envelope is transformed by CycleGAN from emotion to neutral, and WORLD synthesizes the sanitized waveform. Evaluated on RAVDESS with happy and angry speech converted to neutral, the pipeline drops the accuracy of a seven-class emotion classifier by roughly 96%, raises word error rate to 35%, and moves speaker-recognition equal error rate by about 0.12 percentage points. The paper concludes that an edge-side privacy layer can preserve utility for content and identity while suppressing emotional inference.

Load-bearing premise

The argument assumes that the 96% drop in emotion recognition is caused by removing emotion-specific acoustic cues and not by generic degradation from the voice conversion.

Editorial extensions

If this is right

A cloud service receiving the converted audio would no longer be able to estimate a user's emotional state from prosody with the same reliability, since the emotion classifier's accuracy drops by about 96%.
Speech-to-text and speaker verification remain usable on the sanitized signal, with a reported word error rate of 35% and a speaker-recognition error-rate change of about 0.12%.
The same framework is not limited to happy and angry speech: any emotional style can be placed on the source side of the conversion, so the privacy layer can be extended to other paralinguistic attributes.
Deploying this layer on the user's device would let voice assistants keep their cloud-based intelligence while the raw affective signal never leaves the edge.

Reading between the lines

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

If the 96% drop is genuinely emotion-specific rather than a vocoder artifact, the same conversion may also suppress other paralinguistic inferences encoded in prosody, such as stress, confidence, or health cues; that is a testable extension the paper does not run.
A word error rate of 35% may be acceptable for short voice commands or wake-word detection, but not for open-ended dictation; the practical deployment niche is probably command-and-control rather than full transcription.
Privacy here is selective: it makes one inference task fail rather than hiding the whole utterance, so a determined provider could still extract other cues; combining conversion with content filtering or differential privacy would close that gap.

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.

Desk Editor's Note

read the letter

The core idea is sensible and worth taking seriously: use CycleGAN-VC2 and WORLD to convert emotional speech to neutral before it hits a cloud service, so a downstream emotion classifier fails while words and speaker identity survive. That is a genuine application of existing tools to a privacy problem that matters, and the authors had the good sense to evaluate on three tasks (emotion, speech recognition, speaker recognition) and to ship code and results. The 35% WER is reported openly, which is more honest than most papers at this stage.

The problem is that the evaluation only supports half of the claim. The paper shows that an emotion classifier trained on RAVDESS no longer recognizes emotions in converted RAVDESS speech. It does not show that emotional information was removed. The missing neutral-to-neutral control is load-bearing: without passing neutral speech through the same WORLD resynthesis and CycleGAN mapping, the 96% drop could simply mean “the output is so noisy that the classifier collapses,” not “emotion was selectively projected away.” The reported 35% WER provides direct evidence that the pipeline injects heavy artifacts on exactly the files used for evaluation, so the control is not a nicety — it is essential.

Other soft spots: the test set is 40 files, no error bars or statistical tests, and the text does not explain how those 40 files are disjoint from the CycleGAN training set and the emotion classifier’s training set. Since both are RAVDESS-based and likely heavily overlapping, label leakage cannot be ruled out. The speaker recognition result (EER change of ~0.12%) is oddly clean and lacks detail. And the paper’s own language undercuts the utility claim: calling a 35% WER a “minimal” effect on speech recognition is not plausible, even for a lab prototype.

Who gets value from this? Researchers working on privacy-preserving speech processing will find the threat model and the conversion setup a useful starting point, and the released code is a plus. As a submission, it is a typical early workshop paper: promising direction, weak experimental design. A serious referee should not desk-reject it, but should send it back with a clear demand for a neutral-to-neutral control, a larger test set, error bars, and a more careful discussion of what the 35% WER means for the privacy-utility tradeoff. I would not cite the quantitative claims until those are fixed.

Referee Report

3 major / 6 minor

Summary. The paper proposes a privacy-preserving intermediate layer for voice assistants that sanitizes emotional information from speech before forwarding it to cloud services. The pipeline extracts WORLD vocoder features (log F0, spectral envelopes, aperiodicity) and uses a CycleGAN (in the style of CycleGAN-VC2) to convert emotional speech to neutral speech. The authors evaluate the approach on the RAVDESS dataset using three downstream tasks: emotion recognition, speech recognition, and speaker recognition. They report that emotion recognition accuracy drops by approximately 96%, the word error rate (WER) increases to 35%, and speaker recognition degrades only slightly (about 0.12%). The central claim is that the method removes sensitive emotional state while preserving speech content and speaker identity well enough for practical use.

Significance. The problem is well motivated and the manuscript makes a useful contribution by applying neural voice conversion to privacy-preserving speech analysis and by openly providing code and results. The threat model is clearly described, and the choice to evaluate on three standard tasks is sensible. However, the experimental support for the central privacy-utility trade-off is incomplete: the missing neutral-to-neutral control conversion, the unspecified overlap between training and test data, and the high word error rate mean the paper's main conclusion is not yet established. If the authors can address these issues, the approach has potential, but in its current form the evidence does not justify the stated claims.

major comments (3)

[Section 6, Figure 4] The evaluation lacks a control condition in which neutral (natural) speech is passed through the same WORLD/CycleGAN conversion pipeline and then scored by the emotion classifier. Without this control, the reported 96% emotion recognition drop cannot be attributed to emotion-specific sanitization; it could be caused by generic vocoder or conversion artifacts. The 35% word error rate reported in the same section confirms that these artifacts are substantial. Please add a neutral-to-neutral conversion condition and report emotion classification results on original emotional, converted emotional, and converted neutral audio.
[Section 4.2, Section 6] The paper does not specify how the 40 test recordings are chosen or whether they are disjoint from the training sets of the CycleGAN converter and the RAVDESS emotion classifier. Since both models are trained on RAVDESS, overlapping test data would directly inflate the privacy result. The statement in Section 6 that the test audio is "different from the training set" is not backed by any description of the split. Please provide the exact data split, including actor and session identifiers, and report classifier and converter performance on original audio for the same test set.
[Section 6, Figure 4] A 35% word error rate is a severe degradation of speech recognition and is inconsistent with the claim that utility is "minimally" affected. The paper also reports no WER for the original recordings, no error bars, and no statistical tests, and the evaluation is based on only 40 test recordings. The utility side of the privacy-utility trade-off is therefore not established. Please report WER and EER for both original and converted audio with confidence intervals, and discuss whether a 35% WER is acceptable for the intended voice-assistant scenario.

minor comments (6)

[Section 4.1, Section 4.2] The number of emotion classes is inconsistent: Section 4.1 lists calm, happy, sad, angry, fearful, surprise, and disgust (seven emotions excluding neutral), whereas Section 4.2 lists eight labels 0-7 including neutral. Please define the exact label set used in the emotion classifier.
[Section 6] The text "40 emoational recorded from RAVDASS dataset which is diffrent from the training set" contains typos and should be rewritten to describe the test-set construction precisely.
[Section 4.2] Equation (1) includes the hyperparameter lambda, but its value and any sensitivity analysis are not reported. Please state the lambda used in the experiments.
[Section 6] The paper says speaker recognition "accuracy" decreased by about 0.12%, but Section 6 defines speaker recognition performance by equal error rate (EER). Please clarify which metric is being reported.
[References] Reference [5] is a duplicate of reference [4]; both cite VoxCeleb2.
[Figure 4] The figure is referenced in the text, but the caption and axes are not described in sufficient detail. Please describe what each bar represents and the units of the y-axis.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the claim is an experimentally measured privacy-utility tradeoff, with no fitted parameter or self-citation chain reducing the result to its inputs.

full rationale

The paper's central result is an empirical evaluation: emotional RAVDESS utterances are converted to neutral-style speech with a CycleGAN, and the converted files are scored by a separately trained emotion classifier. The 96% drop in emotion recognition is reported as a measured outcome (Section 6), not derived from the CycleGAN objective or from a fitted parameter. The CycleGAN is trained with adversarial and cycle-consistency losses (Eq. 1) to map emotional spectral features to features of normal speech; it is not trained to minimize the emotion classifier's accuracy, and the emotion classifier is an external model (GitHub repository [15]). The shared RAVDESS source is a dataset-overlap concern and the missing neutral-to-neutral control is a genuine attribution threat (generic WORLD/CycleGAN artifacts could depress classifier accuracy), but neither makes the claim circular: the paper does not define the sanitized output as 'what the emotion classifier fails on,' nor does it invoke an author-specific uniqueness theorem. References to CycleGAN-VC2 are independent prior work, not a self-citation chain. Therefore no circular step can be exhibited with the required specificity.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities. Its dependence is on standard WORLD features, an existing CycleGAN-VC2 architecture, and a set of domain assumptions about the sufficiency of these features to carry and then remove emotional information.

free parameters (1)

CycleGAN loss weight lambda = not reported
Lambda controls the relative importance of the adversarial loss and the cycle consistency loss in the CycleGAN objective. The paper does not state the value used, and the conversion quality could depend on this choice.

assumptions (4)

domain assumption WORLD features (F0, spectral envelope, aperiodicity) capture the emotion-relevant acoustic information and are sufficient for emotion conversion.
Section 3.1 and Figure 2: the entire method depends on these three feature sets being the right carrier of emotional content in speech.
domain assumption CycleGAN-VC2 can transform emotional spectral features into neutral ones while preserving linguistic content and speaker identity.
Section 3.2: the paper relies on this capability without a formal or independent evaluation of content preservation beyond the final WER and EER numbers.
domain assumption The accuracy of a single emotion classifier trained on RAVDESS is a valid and sufficient measure of residual emotional privacy.
Section 4: privacy is measured only through this one classifier; the paper does not test against other emotion recognition systems or inference attacks.
domain assumption The 40 test recordings are truly disjoint from the training data of both the converter and the emotion classifier.
Section 6 states the recordings are different from the training set, but the paper does not provide details on speaker or file-level separation, leaving a potential data leakage concern.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Emotionless: Privacy-Preserving Speech Analysis for Voice Assistants." pith.science (2026). https://pith.science/paper/VRCTY2SY

@misc{pith2026190803632,
  author       = {Pith},
  title        = {Pith review of: Emotionless: Privacy-Preserving Speech Analysis for Voice Assistants},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VRCTY2SY}},
  note         = {Machine review of arXiv:1908.03632}
}

read the original abstract

Voice-enabled interactions provide more human-like experiences in many popular IoT systems. Cloud-based speech analysis services extract useful information from voice input using speech recognition techniques. The voice signal is a rich resource that discloses several possible states of a speaker, such as emotional state, confidence and stress levels, physical condition, age, gender, and personal traits. Service providers can build a very accurate profile of a user's demographic category, personal preferences, and may compromise privacy. To address this problem, a privacy-preserving intermediate layer between users and cloud services is proposed to sanitize the voice input. It aims to maintain utility while preserving user privacy. It achieves this by collecting real time speech data and analyzes the signal to ensure privacy protection prior to sharing of this data with services providers. Precisely, the sensitive representations are extracted from the raw signal by using transformation functions and then wrapped it via voice conversion technology. Experimental evaluation based on emotion recognition to assess the efficacy of the proposed method shows that identification of sensitive emotional state of the speaker is reduced by ~96 %.

Figures

Figures reproduced from arXiv: 1908.03632 by the authors.

**Figure 1.** Block diagram of the proposed privacypreserving framework for speech analysis loss expresses the objective of the generators that attempt to fool its corresponding discriminator into being less able to distinguish its generated output from the real one. Cyclic loss calculates the loss of translating a sample from Y to X and then back again to Y. The full objective is: LossFull = LossAdv + λLossCyc (1) where hyperpa… view at source ↗

**Figure 2.** Acoustic features comparison between three [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** A basic architecture of CycleGAN to trans [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Accuracy comparison between three speech [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

FeatureSense: Protecting Speaker Attributes in Always-On Audio Sensing System
cs.SD 2025-05 conditional novelty 6.0 of 10

FeatureSense exposes hand-picked audio features instead of raw audio and introduces the SILI metric, claiming 60.6% lower speaker attribute leakage while keeping sound classification accuracy.

Reference graph

Works this paper leans on

30 extracted references · 27 canonical work pages · cited by 1 Pith paper

[1]

[n. d.]. Cloud Speech-to-Text - Speech Recognition | Cloud Speech-to- Text | Google Cloud. https://cloud.google.com/speech-to-text/

work page
[2]

Efthimios Alepis and Constantinos Patsakis. 2017. Monkey says, mon- key does: security and privacy on voice assistants.IEEE Access 5 (2017), 17841–17851

work page 2017
[3]

D Childers, B Yegnanarayana, and Ke Wu. 1985. Voice conversion: Fac- tors responsible for quality. InICASSP’85. IEEE International Conference on Acoustics, Speech, and Signal Processing , Vol. 10. IEEE, 748–751

work page 1985
[5]

Joon Son Chung, Arsha Nagrani, and Andrew Zisserman. 2018. Vox- celeb2: Deep speaker recognition. arXiv preprint arXiv:1806.05622 (2018)

arXiv 2018
[6]

Carl Doersch and Andrew Zisserman. 2017. Multi-task self-supervised visual learning. In Proceedings of the IEEE International Conference on Computer Vision. 2051–2060

work page 2017
[7]

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2014. Generative adversarial nets. In Advances in neural information process- ing systems. 2672–2680

work page 2014
[8]

Karol Grabowski, Agnieszka Rynkiewicz, Amandine Lassalle, Simon Baron-Cohen, Björn Schuller, Nicholas Cummins, Alice Baird, Justyna Podgórska-Bednarz, Agata Pieniążek, and Izabela Łucka. 2019. Emo- tional expression in psychiatric conditions: New technology for clini- cians. Psychiatry and clinical neurosciences 73 (2019), 50–62

work page 2019
[9]

Zeynep Inanoglu and Steve Young. 2007. A system for transforming the emotion in speech: Combining data-driven conversion techniques for prosody and voice quality. In Eighth Annual Conference of the International Speech Communication Association

work page 2007

Show all 30 references

[10]

Takuhiro Kaneko, Hirokazu Kameoka, Kou Tanaka, and Nobukatsu Hojo. 2019. CycleGAN-VC2: Improved CycleGAN-based Non-parallel Voice Conversion. In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing

2019
[11]

Robert M Krauss, Robin Freyberg, and Ezequiel Morsella. 2002. In- ferring speakersâĂŹ physical attributes from their voices. Journal of Experimental Social Psychology 38, 6 (2002), 618–625

2002
[12]

Steven R Livingstone and Frank A Russo. 2018. The Ryerson Audio- Visual Database of Emotional Speech and Song (RAVDESS): A dynamic, multimodal set of facial and vocal expressions in North American English. PloS one (2018), e0196391

2018
[13]

Ana Ramírez López, Shreyas Seshadri, Lauri Juvela, Okko Räsänen, and Paavo Alku. 2017. Speaking Style Conversion from Normal to Lombard Speech Using a Glottal Vocoder and Bayesian GMMs. 1363–1367

2017
[14]

François Mairesse, Marilyn A Walker, Matthias R Mehl, and Roger K Moore. 2007. Using linguistic cues for the automatic recognition of personality in conversation and text. Journal of artificial intelligence research 30 (2007), 457–500

2007
[15]

Marcogdepinto. 2019. marcogdepinto/Emotion-Classification-Ravdess. https://github.com/marcogdepinto/Emotion-Classification-Ravdess

2019
[16]

Masanori Morise. 2015. CheapTrick, a spectral envelope estimator for high-quality speech synthesis. Speech Communication (2015), 1–7

2015
[17]

Masanori Morise. 2016. D4C, a band-aperiodicity estimator for high- quality speech synthesis. Speech Communication (2016), 57–65

2016
[18]

Masanori Morise, Hideki Kawahara, and Haruhiro Katayose. 2009. Fast and reliable F0 estimation method based on the period extraction of vocal fold vibration of singing voice and speech. In Audio Engineering Society Conference: 35th International Conference: Audio for Games

2009
[19]

Masanori Morise, Fumiya Yokomori, and Kenji Ozawa. 2016. WORLD: a vocoder-based high-quality speech synthesis system for real-time applications. IEICE TRANSACTIONS on Information and Systems (2016), 1877–1884

2016
[20]

Iosif Mporas and Todor Ganchev. 2009. Estimation of unknown speak- erâĂŹs height from speech. International Journal of Speech Technology 12, 4 (2009), 149–160

2009
[21]

Scott R Peppet. 2014. Regulating the internet of things: first steps toward managing discrimination, privacy, security and consent. Tex. L. Rev. 93 (2014), 85

2014
[22]

Jianwei Qian, Haohua Du, Jiahui Hou, Linlin Chen, Taeho Jung, and Xiang-Yang Li. 2018. Hidebehind: Enjoy Voice Input with Voiceprint Unclonability and Anonymity. In Proceedings of the 16th ACM Confer- ence on Embedded Networked Sensor Systems . ACM, 82–94

2018
[23]

2013.Computational paralinguistics: emotion, affect and personality in speech and language processing

Björn Schuller and Anton Batliner. 2013.Computational paralinguistics: emotion, affect and personality in speech and language processing . John Wiley & Sons

2013
[24]

Björn Schuller, Stefan Steidl, Anton Batliner, Alessandro Vinciarelli, Klaus Scherer, Fabien Ringeval, Mohamed Chetouani, Felix Weninger, Florian Eyben, Erik Marchi, et al. 2013. The INTERSPEECH 2013 com- putational paralinguistics challenge: Social signals, conflict, emotion, autism

2013
[25]

Aaron Sell, Gregory A Bryant, Leda Cosmides, John Tooby, Daniel Sznycer, Christopher Von Rueden, Andre Krauss, and Michael Gurven
[26]

George Trigeorgis, Fabien Ringeval, Raymond Brueckner, Erik Marchi, Mihalis A Nicolaou, Björn Schuller, and Stefanos Zafeiriou. 2016. Adieu features? end-to-end speech emotion recognition using a deep convo- lutional recurrent network. In 2016 IEEE international conference on ...

2016
[27]

Zhizheng Wu, Nicholas Evans, Tomi Kinnunen, Junichi Yamagishi, Federico Alegre, and Haizhou Li. 2015. Spoofing and countermeasures for speaker verification: A survey. speech communication (2015), 130– 153

2015
[28]

Weidi Xie, Arsha Nagrani, Joon Son Chung, and Andrew Zisserman
[29]

Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. 2017. Unpaired image-to-image translation using cycle-consistent adver- sarial networks. In Proceedings of the IEEE international conference on computer vision. 2223–2232

2017
[2010]

Proceedings of the Royal Society B: Biological Sciences 277, 1699 (2010), 3509–3518

Adaptations in humans for assessing physical strength from the voice. Proceedings of the Royal Society B: Biological Sciences 277, 1699 (2010), 3509–3518

2010
[2019]

arXiv preprint arXiv:1902.10107 (2019)

Utterance-level Aggregation For Speaker Recognition In The Wild. arXiv preprint arXiv:1902.10107 (2019)

2019 arXiv

Pith tools

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

[1] [1]

[n. d.]. Cloud Speech-to-Text - Speech Recognition | Cloud Speech-to- Text | Google Cloud. https://cloud.google.com/speech-to-text/

work page

[2] [2]

Efthimios Alepis and Constantinos Patsakis. 2017. Monkey says, mon- key does: security and privacy on voice assistants.IEEE Access 5 (2017), 17841–17851

work page 2017

[3] [3]

D Childers, B Yegnanarayana, and Ke Wu. 1985. Voice conversion: Fac- tors responsible for quality. InICASSP’85. IEEE International Conference on Acoustics, Speech, and Signal Processing , Vol. 10. IEEE, 748–751

work page 1985

[4] [5]

Joon Son Chung, Arsha Nagrani, and Andrew Zisserman. 2018. Vox- celeb2: Deep speaker recognition. arXiv preprint arXiv:1806.05622 (2018)

arXiv 2018

[5] [6]

Carl Doersch and Andrew Zisserman. 2017. Multi-task self-supervised visual learning. In Proceedings of the IEEE International Conference on Computer Vision. 2051–2060

work page 2017

[6] [7]

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2014. Generative adversarial nets. In Advances in neural information process- ing systems. 2672–2680

work page 2014

[7] [8]

Karol Grabowski, Agnieszka Rynkiewicz, Amandine Lassalle, Simon Baron-Cohen, Björn Schuller, Nicholas Cummins, Alice Baird, Justyna Podgórska-Bednarz, Agata Pieniążek, and Izabela Łucka. 2019. Emo- tional expression in psychiatric conditions: New technology for clini- cians. Psychiatry and clinical neurosciences 73 (2019), 50–62

work page 2019

[8] [9]

Zeynep Inanoglu and Steve Young. 2007. A system for transforming the emotion in speech: Combining data-driven conversion techniques for prosody and voice quality. In Eighth Annual Conference of the International Speech Communication Association

work page 2007