Pith. sign in

REVIEW 3 major objections 5 minor 162 references

AffectDF: The Most Comprehensive Benchmark for Speech Deepfake Detection against Emotionally Expressive Attacks

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read This paper claims that speech deepfake detectors trained on conventional benchmarks collapse toward random accuracy on emotionally expressive synthetic speech, and that even large-scale emotional training does not reliably fix the collapse.

desk verdict Substantial, well-built emotional deepfake benchmark with a credible core degradation result, but the causal language about emotion overreaches the design; worth serious review with revisions. read the letter →

arxiv 2608.05507 v1 pith:MCS626OU submitted 2026-08-06 eess.AS

Aurosweta Mahapatra , Xiutian Zhao , Shreeram Suresh Chandra , Zihan Zhang , Zongyang Du , Ismail Rasim Ulgen , Kong Aik Lee , Nicholas Andrews
show 2 more authors
Carlos Busso Berrak Sisman
This is my paper · ORCID
classification eess.AS
keywords speechdeepfakedetectionemotionalprosodyspoofingbenchmarkcross-domaingeneralizationlargeaudio-languagemodelsvoiceconversiontext-to-speechactedvsspontaneous
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

Speech deepfake detection systems look strong on standard benchmarks but fail badly when the synthetic speech carries emotion. This paper builds AffectDF, a roughly 260-hour benchmark of synthetic speech spanning 21 generation systems ranging from text-to-speech and voice conversion to large audio-language models, across five emotional states and both acted and spontaneous speech. Evaluating six detector architectures plus three large-audio-language-model detectors, the paper finds that conventional training degrades to near-random performance on AffectDF (RawNet2 at 59.71% EER, AASIST at 56.40%), and that training on AffectDF itself does not consistently transfer to other benchmarks. The paper concludes that current systems rely on domain- and attack-specific cues rather than generalized spoof representations that survive emotional and prosodic variation. A reader should care because the failure mode is precisely the one real attackers would exploit: expressive, naturalistic synthetic speech rather than neutral read utterances.

What carries the argument

The load-bearing object is the AffectDF benchmark itself: a leave-one-factor-varied dataset built from two source corpora, the acted Emotional Speech Dataset (ESD) and the spontaneous MSP-Podcast corpus, with 21 spoofing attacks spanning TTS, VC, EVC, VC+EVC, and LALM-based EVC across the neutral, happy, angry, sad, and surprised states. Its split design is the mechanism that makes the claims testable: train and development partitions use speakers and attack systems disjoint from the test partition, whose 14 unseen attacks include spontaneous-speech attacks A14, A19, and A21 matched to acted attacks A13, A18, and A20 by generation model, so a performance drop on the test set cannot be explained by memorized speakers or seen generators. The released protocol files carry emotion, attack, generation-model, speaker, and source-corpus metadata, which is what allows the emotion-wise, attack-wise, and acted-versus-spontaneous attribution of the failures.

What would settle it

Retrain a strong detector on AffectDF and evaluate it on synthetic speech produced by the same 21 systems from neutral renditions of the same utterances by the same speakers. If error rates on the neutral controls match those on the emotional renditions, the claim that emotional prosody specifically breaks detection would be falsified; if the neutral controls are detected near perfectly, the test would confirm that prosody is the active variable rather than a corpus or generator confound.

Watch

Extended reading notes

Core claim

The paper's central claim is that emotional prosody breaks current speech deepfake detectors. Models trained on conventional benchmarks such as ASVspoof2019 and ASVspoof5 achieve strong in-domain results, but their error rates soar on AffectDF, with the simpler architectures approaching chance (RawNet2 at 59.71% EER, AASIST at 56.40%), and no single emotion is uniformly hardest: the most confusable state shifts with the model and its training data. Surprisingly, supervised training on AffectDF's roughly 87,000 emotional synthetic-speech samples does not consistently improve cross-domain robustness; it often improves scores on smaller emotional benchmarks while severely degrading performance on conventional ones, and the direction of the effect varies by architecture (RawNet2 and the fine-tuned VoxTral improve, while XLSR-Mamba and ProSDD degrade). This leads the authors to state that the bottleneck is not simply the amount of emotional training data but the models' inability to learn spoof representations that generalize across emotional states, attack families, and speaking styles.

Load-bearing premise

The conclusion that emotion causes the failures assumes the ESD and MSP-Podcast emotion labels are valid and that emotional prosody is the main thing varying across datasets; the acted-versus-spontaneous comparison is also confounded by corpus-level differences such as read versus natural speech, recording conditions, and utterance duration, with no neutral-speech control run on the same 21 attacks.

Editorial extensions

If this is right

  • Detectors deployed against emotional synthetic speech in impersonation, fraud calls, or misinformation will operate near chance unless they are trained or designed specifically for emotional and prosodic variability.
  • A low error rate on a small emotional benchmark such as EmoFake does not indicate robustness on broader emotional conditions; evaluation needs attack diversity and speaking-style coverage comparable to AffectDF's.
  • Adding more emotional training data is not by itself the fix, since AffectDF-trained models transfer inconsistently and often sacrifice conventional-benchmark performance.
  • LALM-based detectors, both inference-only and fine-tuned, share the same fragility: prompting alone does not yield robust detection, and supervised fine-tuning transfers poorly across paralinguistic and dataset shifts.

Reading between the lines

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

  • A testable extension would be a neutral-prosody control using the same 21 generators and the same speakers and utterances; if error rates on neutral renditions match those on emotional renditions, the attribution of the collapse to emotion specifically would be weakened in favor of corpus-level confounds that the paper acknowledges in its appendix.
  • The emotion-wise instability (the hardest emotion shifting across models) hints that detectors latch onto prosodic correlates such as pitch, energy, and speaking rate rather than synthesis artifacts, so prosody-conditioned augmentation or prosody-disentangled training may be a more direct remedy than adding emotional data.
  • The observation that the more diverse conventional ASVspoof5 training partially transfers to unseen emotional VC attacks suggests attack diversity, rather than emotion labels themselves, may be the transferable ingredient; this could be tested by ablating AffectDF training data stratified by attack family.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. This paper introduces AffectDF, a large benchmark for emotionally expressive speech deepfake detection, containing roughly 260 hours of speech generated by 21 spoofing attacks spanning TTS, VC, emotional VC, and LALM-based systems, built on acted (ESD) and spontaneous (MSP-Podcast) emotional speech. The authors evaluate RawNet2, AASIST, XLSR-SLS, XLSR-Mamba, ProSDD, and several LALM-based detectors under inference-only and fine-tuned settings, across conventional benchmarks (ASVspoof2019, ASVspoof2021, ASVspoof5) and emotional benchmarks (EmoFake, EmoSpoof-TTS, AffectDF). The main empirical finding is that models trained on conventional benchmarks degrade sharply on AffectDF, sometimes to near-random EER levels, and that training on AffectDF does not consistently improve cross-domain robustness. The paper also analyzes emotion-wise, attack-wise, and acted-versus-spontaneous performance variation. The authors conclude that current SDD systems fail to learn generalized spoof representations under emotional and prosodic variability.

Significance. The dataset is a substantial and potentially valuable community resource: it is larger than prior emotional spoofing datasets, spans a broad range of modern attack families including LALM-based methods, includes both acted and spontaneous emotional speech, and ships detailed protocol metadata. The paper also includes useful checks that strengthen the benchmark's credibility: disjoint speakers and attack systems across splits, documented emotion balance (Table 8), an evaluation of low-level differential cues (Appendix G), and detailed reproducibility settings (Appendix B.3). If the dataset and protocols are released as stated, the empirical evidence of poor cross-domain generalization on AffectDF is a meaningful contribution. However, the causal claim that emotional prosody specifically drives the observed failures is not established by the current experimental design, since the comparisons confound emotion with corpus-level factors such as source corpus, recording conditions, utterance duration, and attack composition. The paper's value as a benchmark does not depend entirely on that causal claim, but the conclusions as written overstate what the design can support.

major comments (3)
  1. [§5.1, Table 2] The section heading 'Emotion Breaks Cross-Domain Generalization' asserts a causal role for emotion, but the design compares full datasets that differ simultaneously in attack composition (LALM, VC, EVC, TTS), source corpus, utterance duration, and speaking style, not only in emotional content. The severe EER degradation on AffectDF is credible evidence of poor generalization, but without a neutral-speech subset generated by the same 21 attack systems, the paper cannot attribute the near-random performance to emotional prosody rather than to other dataset shifts. Please either add such a control or rephrase the claim as a descriptive robustness gap rather than a causal effect of emotion.
  2. [§5.4, Table 5 and Appendix H] The acted-versus-spontaneous comparison is intended to isolate speaking style while keeping the generation method fixed, but the acted and spontaneous attacks also differ by source corpus (ESD vs MSP-Podcast), read versus natural speech, recording conditions, speaker population, and average utterance duration (Table 15 shows spontaneous samples are roughly twice as long). The authors acknowledge these confounds in Appendix H and in the Limitations section, yet RQ3 and Section 5.4 interpret the results as effects of acted versus spontaneous speech. This is a load-bearing inference for the paper's causal narrative. The conclusions should be softened, or the analysis should include a corpus-matched, duration-matched, or neutral-style condition that removes these confounds.
  3. [§4, Tables 2–5] All EER results are single-run point estimates with no confidence intervals, significance tests, or across-seed variation. Several qualitative claims are based on small or irregular differences, such as the cross-training comparisons for ProSDD and the acted/spontaneous overall columns in Table 5. Since training stochasticity alone could explain some of these patterns, the key comparative claims should be supported by multiple runs or a statistical test; otherwise the strength of the conclusions should be reduced to descriptive observations.
minor comments (5)
  1. [Throughout] Model names are inconsistently rendered with extra spaces (e.g., 'CosyV oice', 'V oxtral'), and some reference entries contain typos (e.g., 'JianZhao JianZhao' in the F5-TTS citation); a thorough copyedit is needed.
  2. [Appendix E, Table 12] The model name 'XLSR-MA' is used inconsistently for XLSR-Mamba; please unify the abbreviation with the main text.
  3. [Table 5] The Overall columns appear to be pooled EER values rather than simple averages of the per-attack EERs; the text should state this explicitly, since otherwise the values look arithmetically inconsistent (e.g., Voxtral ASVspoof2019 spontaneous Overall is 45.44 while per-attack values are 24.93, 40.77, and 27.67).
  4. [§3.3] The text says the generation method is kept fixed when comparing acted and spontaneous conditions, but the base corpus also changes from ESD to MSP-Podcast; please qualify this as 'generation method fixed, source corpus changed.'
  5. [Appendix I] The SER analysis relies on emotion2vec+ fine-tuned on EmoBox, which includes ESD and MSP-Podcast; the authors note this caveat, but the confusion matrix in Figure 3 should also be cited in the main text when discussing the emotional expressiveness of AffectDF.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: AffectDF is an empirical benchmark paper whose degradation results are measurements against external detectors and attack systems, not derivations from fitted inputs.

full rationale

The paper's central claims are empirical evaluations: models trained on conventional benchmarks are tested on a newly constructed dataset and their EERs are reported. No parameter is fitted to a subset of the evaluation data and then renamed as a prediction. The dataset itself is constructed from external corpora (ESD, MSP-Podcast) and external generation systems (TTS, VC, EVC, LALM), and the benchmark results are measured against standard detection architectures. Self-citations, such as EmoSpoof-TTS, ProSDD, and the ESN steering framework, are used as baselines or attack-generation tools, not as load-bearing justifications of the main empirical findings. The acknowledged confounds in the acted-versus-spontaneous comparison (Appendix H and the Limitations section) weaken the causal claim that emotional prosody specifically drives the degradation, but confounding is a validity concern, not circularity: the paper does not define its outcome in terms of its input, nor does it fit a parameter that forces the reported EERs. The Appendix I SER check using emotion2vec+ is explicitly labeled as an approximate expressiveness estimate and is not used to derive the central robustness results. Therefore, no circular step can be exhibited from the paper's own equations or construction, and the appropriate finding is no significant circularity.

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

No free parameters are fitted in this benchmark paper. The evaluation rests on external corpus labels, split hygiene, and the standard EER protocol. No new theoretical entities such as particles, forces, dimensions, or conserved quantities are introduced.

assumptions (4)
  • domain assumption ESD and MSP-Podcast emotion labels and acted/spontaneous designations accurately represent the intended emotional conditions.
    Used in Section 3.2.1 to construct AffectDF and in Sections 5.2 and 5.4 to interpret emotion-wise and style-wise EER differences.
  • domain assumption Generated samples contain no undisclosed leakage through reference audio or overlap with bona fide source utterances.
    Appendix A.2 states that reference audio was held out; the benchmark conclusions assume this holdout is complete and that the released splits prevent leakage.
  • domain assumption Disjoint speakers and disjoint attack systems between train and test are sufficient to measure cross-domain generalization.
    Section 3.2.3 defines the split policy, and the cross-domain conclusions in Section 5.1 rely on this separation.
  • standard math EER computed on pooled spoof/real scores is a valid and sufficient robustness metric for benchmark ranking.
    Section 4 states EER as the primary metric, but no confidence intervals or statistical tests accompany the reported values.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AffectDF: The Most Comprehensive Benchmark for Speech Deepfake Detection against Emotionally Expressive Attacks." pith.science (2026). https://pith.science/paper/MCS626OU

@misc{pith2026260805507,
  author       = {Pith},
  title        = {Pith review of: AffectDF: The Most Comprehensive Benchmark for Speech Deepfake Detection against Emotionally Expressive Attacks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MCS626OU}},
  note         = {Machine review of arXiv:2608.05507}
}
read the original abstract

Speech deepfake detection (SDD) systems achieve strong performance on conventional benchmarks; however, existing datasets provide limited coverage of emotionally expressive and recent large audio-language model (LALM)-based attacks. Existing emotional spoofing datasets are also limited in scale and attack diversity, typically covering only voice conversion (VC) or text-to-speech (TTS) attacks. We introduce AffectDF, the most comprehensive benchmark for emotionally expressive speech deepfakes, spanning TTS, VC, emotional VC, and LALM-based spoofing attacks across both acted and spontaneous emotional speech. AffectDF contains approximately 260 hours of speech generated using 21 spoofing attacks across five emotional states. We benchmark state-of-the-art SDD systems under conventional and emotional spoofing conditions, including LALM-based detectors evaluated with both inference-only prompting and supervised fine-tuning. Our experiments reveal severe robustness degradation when models trained on conventional benchmarks are evaluated on AffectDF, with several systems approaching near-random performance. Surprisingly, even large-scale emotional training does not consistently improve cross-domain robustness, indicating that current SDD systems fail to learn generalized spoof representations under emotional and prosodic variability. Robustness further varies substantially across emotional states, attack families, and acted vs spontaneous emotional speech conditions. These findings expose fundamental limitations of current SDD systems and establish AffectDF as a benchmark for developing more robust spoof detection models.

Figures

Figures reproduced from arXiv: 2608.05507 by the authors.

Figure 1
Figure 1. Overview of AffectDF benchmark, including [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Normalized emotion distribution from Emotion2Vec+-large predictions across AffectDF and conventional [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Emotion confusion matrix for AffectDF using [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

162 extracted references · 32 canonical work pages

  1. [1]

    Busso and R

    C. Busso and R. Lotfian and K. Sridhar and A.N. Salman and W.-C. Lin and L. Goncalves and S. Parthasarathy and A. The. IEEE Transactions on Affective Computing , volume =. 2026 , pages =

    2026

  2. [2]

    Artificial Intelligence Review , volume=

    Battling voice spoofing: a review, comparative analysis, and generalizability evaluation of state-of-the-art voice spoofing counter measures , author=. Artificial Intelligence Review , volume=. 2023 , publisher=

    2023

  3. [3]

    arXiv preprint arXiv:2509.10781 , year=

    Emoanti: audio anti-deepfake with refined emotion-guided representations , author=. arXiv preprint arXiv:2509.10781 , year=

    arXiv

  4. [4]

    Applied intelligence , volume=

    Deepfakes generation and detection: State-of-the-art, open challenges, countermeasures, and way forward , author=. Applied intelligence , volume=. 2023 , publisher=

    2023

  5. [5]

    The 2024 ACM Conference on Fairness, Accountability, and Transparency , pages=

    Not My Voice! A Taxonomy of Ethical and Safety Harms of Speech Generators , author=. The 2024 ACM Conference on Fairness, Accountability, and Transparency , pages=

    2024

  6. [6]

    End-to-End anti-spoofing with RawNet2 , year=

    Tak, Hemlata and Patino, Jose and Todisco, Massimiliano and Nautsch, Andreas and Evans, Nicholas and Larcher, Anthony , booktitle=. End-to-End anti-spoofing with RawNet2 , year=

  7. [7]

    AASIST: Audio Anti-Spoofing Using Integrated Spectro-Temporal Graph Attention Networks , year=

    Jung, Jee-weon and Heo, Hee-Soo and Tak, Hemlata and Shim, Hye-jin and Chung, Joon Son and Lee, Bong-Jin and Yu, Ha-Jin and Evans, Nicholas , booktitle=. AASIST: Audio Anti-Spoofing Using Integrated Spectro-Temporal Graph Attention Networks , year=

  8. [8]

    ASVspoof 2019: A large-scale public database of synthesized, converted and replayed speech , journal =

    Xin Wang and Junichi Yamagishi and Massimiliano Todisco and Héctor Delgado and Andreas Nautsch and Nicholas Evans and Md Sahidullah and Ville Vestman and Tomi Kinnunen and Kong Aik Lee and Lauri Juvela and Paavo Alku and Yu-Huai Peng and Hsin-Te Hwang and Yu Tsao and Hsin-Min Wang and Sébastien Le Maguer and Markus Becker and Fergus Henderson and Rob Clar...

    work page arXiv 2019

Show all 162 references
  1. [9]

    ASVspoof 2021: Towards Spoofed and Deepfake Speech Detection in the Wild , year=

    Liu, Xuechen and Wang, Xin and Sahidullah, Md and Patino, Jose and Delgado, Héctor and Kinnunen, Tomi and Todisco, Massimiliano and Yamagishi, Junichi and Evans, Nicholas and Nautsch, Andreas and Lee, Kong Aik , journal=. ASVspoof 2021: Towards Spoofed and Deepfake Speech Dete...

    2021

  2. [10]

    StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models , url =

    Li, Yinghao Aaron and Han, Cong and Raghavan, Vinay and Mischler, Gavin and Mesgarani, Nima , booktitle =. StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models , url =

  3. [11]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    F5-tts: A fairytaler that fakes fluent and faithful speech with flow matching , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  4. [12]

    arXiv preprint arXiv:2407.05407 , year=

    Cosyvoice: A scalable multilingual zero-shot text-to-speech synthesizer based on supervised semantic tokens , author=. arXiv preprint arXiv:2407.05407 , year=

    arXiv

  5. [13]

    arXiv preprint arXiv:2408.08739 , year=

    ASVspoof 5: Crowdsourced speech data, deepfakes, and adversarial attacks at scale , author=. arXiv preprint arXiv:2408.08739 , year=

    arXiv

  6. [14]

    China National Conference on Chinese Computational Linguistics , pages=

    EmoFake: An initial dataset for emotion fake audio detection , author=. China National Conference on Chinese Computational Linguistics , pages=. 2024 , organization=

    2024

  7. [15]

    ADD 2022: the first Audio Deep Synthesis Detection Challenge , year=

    Yi, Jiangyan and Fu, Ruibo and Tao, Jianhua and Nie, Shuai and Ma, Haoxin and Wang, Chenglong and Wang, Tao and Tian, Zhengkun and Bai, Ye and Fan, Cunhang and Liang, Shan and Wang, Shiming and Zhang, Shuai and Yan, Xinrui and Xu, Le and Wen, Zhengqi and Li, Haizhou , booktitl...

    2022

  8. [16]

    arXiv preprint arXiv:2305.13774 , year=

    Add 2023: the second audio deepfake detection challenge , author=. arXiv preprint arXiv:2305.13774 , year=

    2023 arXiv

  9. [17]

    Training , volume=

    ASVspoof 2015: Automatic speaker verification spoofing and countermeasures challenge evaluation plan , author=. Training , volume=

    2015

  10. [18]

    Seen and Unseen Emotional Style Transfer for Voice Conversion with A New Emotional Speech Dataset , year=

    Zhou, Kun and Sisman, Berrak and Liu, Rui and Li, Haizhou , booktitle=. Seen and Unseen Emotional Style Transfer for Voice Conversion with A New Emotional Speech Dataset , year=

  11. [19]

    The Twelfth International Conference on Learning Representations , year=

    Mega-TTS 2: Boosting Prompting Mechanisms for Zero-Shot Speech Synthesis , author=. The Twelfth International Conference on Learning Representations , year=

  12. [20]

    ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Qi-tts: Questioning intonation control for emotional speech synthesis , author=. ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2023 , organization=

    2023

  13. [21]

    Advances in Neural Information Processing Systems , volume=

    Generspeech: Towards style transfer for generalizable out-of-domain text-to-speech , author=. Advances in Neural Information Processing Systems , volume=

  14. [22]

    Development , volume=

    Odyssey 2024-Speech Emotion Recognition Challenge: Dataset, Baseline Framework, and Results , author=. Development , volume=

    2024

  15. [23]

    Interspeech 2022 , year=

    Exploiting co-occurrence frequency of emotions in perceptual evaluations to train a speech emotion classifier , author=. Interspeech 2022 , year=

    2022

  16. [24]

    arXiv preprint arXiv:2009.09637 , year=

    Light convolutional neural network with feature genuinization for detection of synthetic speech attacks , author=. arXiv preprint arXiv:2009.09637 , year=

    2009 arXiv

  17. [25]

    arXiv preprint arXiv:1904.01120 , year=

    ASSERT: Anti-spoofing with squeeze-excitation and residual networks , author=. arXiv preprint arXiv:1904.01120 , year=

    1904 arXiv

  18. [26]

    2023 18th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP) , pages=

    Spoof Detection using Voice Contribution on LFCC features and ResNet-34 , author=. 2023 18th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP) , pages=. 2023 , organization=

    2023

  19. [27]

    arXiv preprint arXiv:2509.09791 , year=

    The msp-podcast corpus , author=. arXiv preprint arXiv:2509.09791 , year=

    arXiv

  20. [28]

    EMOQ-TTS: Emotion Intensity Quantization for Fine-Grained Controllable Emotional Text-to-Speech , year=

    Im, Chae-Bin and Lee, Sang-Hoon and Kim, Seung-Bin and Lee, Seong-Whan , booktitle=. EMOQ-TTS: Emotion Intensity Quantization for Fine-Grained Controllable Emotional Text-to-Speech , year=

  21. [29]

    Emodiff: Intensity Controllable Emotional Text-to-Speech with Soft-Label Guidance , year=

    Guo, Yiwei and Du, Chenpeng and Chen, Xie and Yu, Kai , booktitle=. Emodiff: Intensity Controllable Emotional Text-to-Speech with Soft-Label Guidance , year=

  22. [30]

    ED-TTS: Multi-Scale Emotion Modeling Using Cross-Domain Emotion Diarization for Emotional Speech Synthesis , year=

    Tang, Haobin and Zhang, Xulong and Cheng, Ning and Xiao, Jing and Wang, Jianzong , booktitle=. ED-TTS: Multi-Scale Emotion Modeling Using Cross-Domain Emotion Diarization for Emotional Speech Synthesis , year=

  23. [31]

    Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers , year=

    Chen, Sanyuan and Wang, Chengyi and Wu, Yu and Zhang, Ziqiang and Zhou, Long and Liu, Shujie and Chen, Zhuo and Liu, Yanqing and Wang, Huaming and Li, Jinyu and He, Lei and Zhao, Sheng and Wei, Furu , journal=. Neural Codec Language Models are Zero-Shot Text to Speech Synthesi...

  24. [32]

    Advances in Neural Information Processing Systems , editor=

    HierSpeech: Bridging the Gap between Text and Speech by Hierarchical Variational Inference using Self-supervised Representations for Speech Synthesis , author=. Advances in Neural Information Processing Systems , editor=. 2022 , url=

    2022

  25. [33]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

  26. [34]

    Laugh Now Cry Later: Controlling Time-Varying Emotional States of Flow-Matching-Based Zero-Shot Text-To-Speech , year=

    Wu, Haibin and Wang, Xiaofei and Eskimez, Sefik Emre and Thakker, Manthan and Tompkins, Daniel and Tsai, Chung-Hsien and Li, Canrun and Xiao, Zhen and Zhao, Sheng and Li, Jinyu and Kanda, Naoyuki , booktitle=. Laugh Now Cry Later: Controlling Time-Varying Emotional States of F...

  27. [35]

    ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    PromptTTS++: Controlling speaker identity in prompt-based text-to-speech using natural language descriptions , author=. ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2024 , organization=

    2024

  28. [36]

    Proceedings of the 32nd ACM International Conference on Multimedia , pages=

    Audio deepfake detection with self-supervised xls-r and sls classifier , author=. Proceedings of the 32nd ACM International Conference on Multimedia , pages=

  29. [37]

    arXiv preprint arXiv:2202.12233 , year=

    Automatic speaker verification spoofing and deepfake detection using wav2vec 2.0 and data augmentation , author=. arXiv preprint arXiv:2202.12233 , year=

    arXiv

  30. [38]

    arXiv preprint arXiv:2111.09296 , year=

    XLS-R: Self-supervised cross-lingual speech representation learning at scale , author=. arXiv preprint arXiv:2111.09296 , year=

    arXiv

  31. [39]

    IEEE Journal of Selected Topics in Signal Processing , volume=

    Wavlm: Large-scale self-supervised pre-training for full stack speech processing , author=. IEEE Journal of Selected Topics in Signal Processing , volume=. 2022 , publisher=

    2022

  32. [40]

    Advances in neural information processing systems , volume=

    wav2vec 2.0: A framework for self-supervised learning of speech representations , author=. Advances in neural information processing systems , volume=

  33. [41]

    IEEE/ACM transactions on audio, speech, and language processing , volume=

    Hubert: Self-supervised speech representation learning by masked prediction of hidden units , author=. IEEE/ACM transactions on audio, speech, and language processing , volume=. 2021 , publisher=

    2021

  34. [42]

    arXiv preprint arXiv:2502.14726 , year=

    Pitch Imperfect: Detecting Audio Deepfakes Through Acoustic Prosodic Analysis , author=. arXiv preprint arXiv:2502.14726 , year=

    arXiv

  35. [43]

    Yet Another Algorithm for Pitch Tracking (YAAPT) , author=

  36. [44]

    , author=

    Harvest: A High-Performance Fundamental Frequency Estimator from Speech Signals. , author=. INTERSPEECH , pages=

  37. [45]

    EURASIP Journal on Audio, Speech, and Music Processing , volume=

    Deep learning-based expressive speech synthesis: a systematic review of approaches, challenges, and resources , author=. EURASIP Journal on Audio, Speech, and Music Processing , volume=. 2024 , publisher=

    2024

  38. [46]

    arXiv preprint arXiv:2504.19197 , year=

    Generative Adversarial Network based Voice Conversion: Techniques, Challenges, and Recent Advancements , author=. arXiv preprint arXiv:2504.19197 , year=

    arXiv

  39. [47]

    Frontiers in signal processing , volume=

    Reimagining speech: a scoping review of deep learning-based methods for non-parallel voice conversion , author=. Frontiers in signal processing , volume=. 2024 , publisher=

    2024

  40. [48]

    IEICE TRANSACTIONS on Information and Systems , volume=

    WORLD: a vocoder-based high-quality speech synthesis system for real-time applications , author=. IEICE TRANSACTIONS on Information and Systems , volume=. 2016 , publisher=

    2016

  41. [49]

    IEEE/ACM Transactions on Audio, Speech, and Language Processing , year=

    Instructtts: Modelling expressive tts in discrete latent space with natural language style prompt , author=. IEEE/ACM Transactions on Audio, Speech, and Language Processing , year=

  42. [50]

    Reinforcement Learning for Emotional Text-to-Speech Synthesis with Improved Emotion Discriminability , author=. Proc. Interspeech 2021 , pages=

    2021

  43. [51]

    arXiv preprint arXiv:2111.07402 , year=

    Textless speech emotion conversion using discrete and decomposed representations , author=. arXiv preprint arXiv:2111.07402 , year=

    arXiv

  44. [52]

    IEEE/ACM Transactions on Audio, Speech, and Language Processing , year=

    PE-wav2vec: A prosody-enhanced speech model for self-supervised prosody learning in TTS , author=. IEEE/ACM Transactions on Audio, Speech, and Language Processing , year=

  45. [53]

    XLSR-Mamba: A Dual-Column Bidirectional State Space Model for Spoofing Attack Detection , year=

    Xiao, Yang and Das, Rohan Kumar , journal=. XLSR-Mamba: A Dual-Column Bidirectional State Space Model for Spoofing Attack Detection , year=

  46. [54]

    Can Emotion Fool Anti-spoofing? , author=. Proc. Interspeech 2025 , pages=

    2025

  47. [55]

    Proceedings of the 15th Biannual Conference of the Italian SIGCHI Chapter , pages=

    How do users perceive deepfake personas? Investigating the deepfake user perception and its implications for human-computer interaction , author=. Proceedings of the 15th Biannual Conference of the Italian SIGCHI Chapter , pages=

  48. [56]

    Better Be Computer or I'm Dumb

    " Better Be Computer or I'm Dumb": A Large-Scale Evaluation of Humans as Audio Deepfake Detectors , author=. Proceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security , pages=

    2024

  49. [57]

    ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Rawboost: A raw data boosting and augmentation method applied to automatic speaker verification anti-spoofing , author=. ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2022 , organization=

    2022

  50. [58]

    International conference on machine learning , pages=

    Robust speech recognition via large-scale weak supervision , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  51. [59]

    ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Mixture of experts fusion for fake audio detection using frozen wav2vec 2.0 , author=. ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2025 , organization=

    2025

  52. [60]

    Machine learning , volume=

    Multitask learning , author=. Machine learning , volume=. 1997 , publisher=

    1997

  53. [61]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    End-to-end multi-task learning with attention , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  54. [62]

    , author=

    Speech emotion recognition with multi-task learning. , author=. Interspeech , volume=. 2021 , organization=

    2021

  55. [63]

    National Science Review , volume=

    An overview of multi-task learning , author=. National Science Review , volume=. 2018 , publisher=

    2018

  56. [64]

    arXiv preprint arXiv:2509.08470 , year=

    Joint Learning using Mixture-of-Expert-Based Representation for Enhanced Speech Generation and Robust Emotion Recognition , author=. arXiv preprint arXiv:2509.08470 , year=

    arXiv

  57. [65]

    ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Multi-task learning improves synthetic speech detection , author=. ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2022 , organization=

    2022

  58. [66]

    arXiv preprint arXiv:1412.3555 , year=

    Empirical evaluation of gated recurrent neural networks on sequence modeling , author=. arXiv preprint arXiv:1412.3555 , year=

    arXiv

  59. [67]

    2021 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , pages=

    Expressive voice conversion: A joint framework for speaker identity and emotional style transfer , author=. 2021 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , pages=. 2021 , organization=

    2021

  60. [68]

    ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Expressive-vc: Highly expressive voice conversion with attention fusion of bottleneck and perturbation features , author=. ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2023 , organization=

    2023

  61. [69]

    Proceedings of the 31st ACM International Conference on Multimedia , pages=

    Pmvc: Data augmentation-based prosody modeling for expressive voice conversion , author=. Proceedings of the 31st ACM International Conference on Multimedia , pages=

  62. [70]

    IEEE Transactions on Information Forensics and Security , volume=

    Unsupervised domain adaptation for face anti-spoofing , author=. IEEE Transactions on Information Forensics and Security , volume=. 2018 , publisher=

    2018

  63. [71]

    arXiv preprint arXiv:2509.21676 , year=

    HuLA: Prosody-Aware Anti-Spoofing with Multi-Task Learning for Expressive and Emotional Synthetic Speech , author=. arXiv preprint arXiv:2509.21676 , year=

  64. [72]

    IEEE Transactions on Affective Computing , volume=

    Emotion intensity and its control for emotional voice conversion , author=. IEEE Transactions on Affective Computing , volume=. 2022 , publisher=

    2022

  65. [73]

    ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Gr0: Self-supervised global representation learning for zero-shot voice conversion , author=. ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2024 , organization=

    2024

  66. [74]

    arXiv preprint arXiv:2311.04693 , year=

    Diff-HierVC: Diffusion-based hierarchical voice conversion with robust pitch generation and masked prior for zero-shot speaker adaptation , author=. arXiv preprint arXiv:2311.04693 , year=

    arXiv

  67. [75]

    2024 IEEE Spoken Language Technology Workshop (SLT) , pages=

    Language-Independent Prosody-Enhanced Speech Representations For Multilingual Speech Synthesis , author=. 2024 IEEE Spoken Language Technology Workshop (SLT) , pages=. 2024 , organization=

    2024

  68. [76]

    ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Towards natural and controllable cross-lingual voice conversion based on neural tts model and phonetic posteriorgram , author=. ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2021 , organization=

    2021

  69. [77]

    Language, Cognition and Neuroscience , volume=

    Prosody in context: A review , author=. Language, Cognition and Neuroscience , volume=. 2015 , publisher=

    2015

  70. [78]

    Studies in second language acquisition , volume=

    Learning second language suprasegmentals: Effect of L2 experience on prosody and fluency characteristics of L2 speech , author=. Studies in second language acquisition , volume=. 2006 , publisher=

    2006

  71. [79]

    and Li, Haizhou , journal=

    Zhou, Kun and Sisman, Berrak and Rana, Rajib and Schuller, Björn W. and Li, Haizhou , journal=. Emotion Intensity and its Control for Emotional Voice Conversion , year=

  72. [80]

    Librispeech: An ASR corpus based on public domain audio books , year=

    Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev , booktitle=. Librispeech: An ASR corpus based on public domain audio books , year=

  73. [81]

    International Conference on Pattern Recognition , pages=

    Combining automatic speaker verification and prosody analysis for synthetic speech detection , author=. International Conference on Pattern Recognition , pages=. 2022 , organization=

    2022

  74. [82]

    ICASSP 2022-2022 IEEE international conference on acoustics, speech and signal processing (ICASSP) , pages=

    Deepfake speech detection through emotion recognition: a semantic approach , author=. ICASSP 2022-2022 IEEE international conference on acoustics, speech and signal processing (ICASSP) , pages=. 2022 , organization=

    2022

  75. [83]

    2020 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) , pages=

    Spectrum and prosody conversion for cross-lingual voice conversion with cyclegan , author=. 2020 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) , pages=. 2020 , organization=

    2020

  76. [84]

    METTS: Multilingual Emotional Text-to-Speech by Cross-Speaker and Cross-Lingual Emotion Transfer , year=

    Zhu, Xinfa and Lei, Yi and Li, Tao and Zhang, Yongmao and Zhou, Hongbin and Lu, Heng and Xie, Lei , journal=. METTS: Multilingual Emotional Text-to-Speech by Cross-Speaker and Cross-Lingual Emotion Transfer , year=

  77. [85]

    arXiv preprint arXiv:2509.00186 , year=

    Generalizable Audio Spoofing Detection using Non-Semantic Representations , author=. arXiv preprint arXiv:2509.00186 , year=

    arXiv

  78. [86]

    IEEE Transactions on Affective Computing , volume=

    Speech synthesis with mixed emotions , author=. IEEE Transactions on Affective Computing , volume=. 2022 , publisher=

    2022

  79. [87]

    2024 International Joint Conference on Neural Networks (IJCNN) , pages=

    Mlaad: The multi-language audio anti-spoofing dataset , author=. 2024 International Joint Conference on Neural Networks (IJCNN) , pages=. 2024 , organization=

    2024

  80. [88]

    arXiv preprint arXiv:2004.06422 , year=

    An explainability study of the constant Q cepstral coefficient spoofing countermeasure for automatic speaker verification , author=. arXiv preprint arXiv:2004.06422 , year=

    2004 arXiv

  81. [89]

    EURASIP Journal on Information Security , volume=

    Synthetic speech detection through short-term and long-term prediction traces , author=. EURASIP Journal on Information Security , volume=. 2021 , publisher=

    2021

  82. [90]

    IEEE/ACM Transactions on Audio, Speech, and Language Processing , volume=

    The impact of silence on speech anti-spoofing , author=. IEEE/ACM Transactions on Audio, Speech, and Language Processing , volume=. 2023 , publisher=

    2023

  83. [91]

    arXiv preprint arXiv:2106.12914 , year=

    Speech is silver, silence is golden: What do ASVspoof-trained models really learn? , author=. arXiv preprint arXiv:2106.12914 , year=

    arXiv

  84. [92]

    Feature Genuinization based Residual Squeeze-and-Excitation for Audio Anti-Spoofing in Sound AI , year=

    Ray, Ruchira and Karthik, Sanka and Mathur, Vinayak and Kumar, Prashant and G, Maragatham and Tiwari, Sourabh and Shankarappa, Rashmi T , booktitle=. Feature Genuinization based Residual Squeeze-and-Excitation for Audio Anti-Spoofing in Sound AI , year=

  85. [93]

    PloS one , volume=

    Mixed emotions and coping: The benefits of secondary emotions , author=. PloS one , volume=. 2014 , publisher=

    2014

  86. [94]

    International Conference on Machine Learning , year=

    Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech , author=. International Conference on Machine Learning , year=

  87. [95]

    arXiv preprint arXiv:2406.04904 , year=

    Xtts: a massively multilingual zero-shot text-to-speech model , author=. arXiv preprint arXiv:2406.04904 , year=

    arXiv

  88. [96]

    ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Fastpitch: Parallel text-to-speech with pitch prediction , author=. ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2021 , organization=

    2021

  89. [97]

    arXiv preprint arXiv:2107.10394 , year=

    Starganv2-vc: A diverse, unsupervised, non-parallel framework for natural-sounding voice conversion , author=. arXiv preprint arXiv:2107.10394 , year=

    arXiv

  90. [98]

    arXiv preprint arXiv:2109.13821 , year=

    Diffusion-based voice conversion with fast maximum likelihood sampling scheme , author=. arXiv preprint arXiv:2109.13821 , year=

    arXiv

  91. [99]

    , author=

    Voice Conversion Using Speech-to-Speech Neuro-Style Transfer. , author=. Interspeech , pages=

  92. [100]

    Advances in Neural Information Processing Systems , volume=

    Glow-tts: A generative flow for text-to-speech via monotonic alignment search , author=. Advances in Neural Information Processing Systems , volume=

  93. [101]

    International conference on machine learning , pages=

    Grad-tts: A diffusion probabilistic model for text-to-speech , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  94. [102]

    arXiv preprint arXiv:2406.03512 , year=

    Harder or different? understanding generalization of audio deepfake detection , author=. arXiv preprint arXiv:2406.03512 , year=

    arXiv

  95. [103]

    arXiv preprint arXiv:2305.13700 , year=

    Detection of cross-dataset fake audio based on prosodic and pronunciation features , author=. arXiv preprint arXiv:2305.13700 , year=

    arXiv

  96. [104]

    2023 , booktitle =

    Towards Cross-Language Prosody Transfer for Dialog , author =. 2023 , booktitle =. doi:10.21437/Interspeech.2023-1152 , issn =

    2023 doi

  97. [105]

    Dubbing in Practice: A Large Scale Study of Human Localization With Insights for Automatic Dubbing

    Brannon, William and Virkar, Yogesh and Thompson, Brian. Dubbing in Practice: A Large Scale Study of Human Localization With Insights for Automatic Dubbing. Transactions of the Association for Computational Linguistics. 2023. doi:10.1162/tacl_a_00551

    2023 doi

  98. [106]

    HABLA: A dataset of Latin American Spanish accents for voice anti-spoofing , author=. Proc. Interspeech , volume=

  99. [107]

    arXiv preprint arXiv:2412.17924 , year=

    Are audio DeepFake detection models polyglots? , author=. arXiv preprint arXiv:2412.17924 , year=

    arXiv

  100. [108]

    Multilingual Audio Deepfakes Dataset for Robust and Generalizable Detection , author=

  101. [109]

    arXiv preprint arXiv:2203.16263 , year=

    Does audio deepfake detection generalize? , author=. arXiv preprint arXiv:2203.16263 , year=

  102. [110]

    Subjective Perception and Objective Evaluation of Speech Naturalness for Deepfake Detection , year=

    Zhang, Siyu and Peng, Liujiahao and Xie, Lingyun and Zhao, Zhijun , booktitle=. Subjective Perception and Objective Evaluation of Speech Naturalness for Deepfake Detection , year=

  103. [111]

    Proceedings of the speech prosody 2002 conference , pages=

    Variation adds to prosodic typology , author=. Proceedings of the speech prosody 2002 conference , pages=. 2002 , organization=

    2002

  104. [112]

    Investigating voiced and unvoiced regions of speech for audio deepfake detection , year=

    Sivaraman, Ganesh and Tak, Hemlata and Khoury, Elie , booktitle=. Investigating voiced and unvoiced regions of speech for audio deepfake detection , year=

  105. [113]

    Advances in Neural Information Processing Systems , volume=

    Slim: Style-linguistics mismatch model for generalized audio deepfake detection , author=. Advances in Neural Information Processing Systems , volume=

  106. [114]

    arXiv preprint arXiv:2506.02584 , year=

    Prosodic Structure Beyond Lexical Content: A Study of Self-Supervised Learning , author=. arXiv preprint arXiv:2506.02584 , year=

    arXiv

  107. [115]

    ACM Computing Surveys , volume=

    A survey on speech deepfake detection , author=. ACM Computing Surveys , volume=. 2025 , publisher=

    2025

  108. [116]

    arXiv preprint arXiv:2005.07143 , year=

    Ecapa-tdnn: Emphasized channel attention, propagation and aggregation in tdnn based speaker verification , author=. arXiv preprint arXiv:2005.07143 , year=

    2005 arXiv

  109. [117]

    Detection of Cross-Dataset Fake Audio Based on Prosodic and Pronunciation Features , author=. Proc. Interspeech 2023 , pages=

    2023

  110. [118]

    Finding the Human Voice in AI: Insights on the Perception of AI-Voice Clones from Naturalness and Similarity Ratings , author=. Proc. Interspeech 2025 , pages=

    2025

  111. [119]

    2025 IEEE International Conference on Cyber Security and Resilience (CSR) , pages=

    Audio-deepfake: Generation Methods, Legitimate Applications and the Potential for Misuse , author=. 2025 IEEE International Conference on Cyber Security and Resilience (CSR) , pages=. 2025 , organization=

    2025

  112. [120]

    EnvFake: An Initial Environmental-Fake Audio Dataset for Scene-Consistency Detection , year=

    Cheng, Hannan and Li, Kangyue and Ye, Long and Wang, Jingling , booktitle=. EnvFake: An Initial Environmental-Fake Audio Dataset for Scene-Consistency Detection , year=

  113. [121]

    arXiv preprint arXiv:2508.20983 , year=

    Multilingual Dataset Integration Strategies for Robust Audio Deepfake Detection: A SAFE Challenge System , author=. arXiv preprint arXiv:2508.20983 , year=

  114. [122]

    2024 IEEE 14th International Symposium on Chinese Spoken Language Processing (ISCSLP) , pages=

    Madd: A multi-lingual multi-speaker audio deepfake detection dataset , author=. 2024 IEEE 14th International Symposium on Chinese Spoken Language Processing (ISCSLP) , pages=. 2024 , organization=

    2024

  115. [123]

    , author=

    FoR: A Dataset for Synthetic Speech Detection. , author=. SpeD , pages=

  116. [124]

    IEEE Open Journal of Signal Processing , year=

    Spoofceleb: Speech deepfake detection and sasv in the wild , author=. IEEE Open Journal of Signal Processing , year=

  117. [125]

    arXiv preprint arXiv:2508.10559 , year=

    Fake Speech Wild: Detecting Deepfake Speech on Social Media Platform , author=. arXiv preprint arXiv:2508.10559 , year=

    arXiv

  118. [126]

    arXiv preprint arXiv:2604.13229 , year=

    ProSDD: Learning Prosodic Representations for Speech Deepfake Detection against Expressive and Emotional Attacks , author=. arXiv preprint arXiv:2604.13229 , year=

    arXiv

  119. [127]

    arXiv preprint arXiv:2512.08403 , year=

    DFALLM: Achieving Generalizable Multitask Deepfake Detection by Optimizing Audio LLM Components , author=. arXiv preprint arXiv:2512.08403 , year=

  120. [128]

    Proceedings of the 1st on Deepfake Forensics Workshop: Detection, Attribution, Recognition, and Adversarial Challenges in the Era of AI-Generated Media , pages=

    SASDN: A Generalizable and Minimal-Intervention LLM-Integrated Framework for Continual Adaptation in Spoofed Speech Detection , author=. Proceedings of the 1st on Deepfake Forensics Workshop: Detection, Attribution, Recognition, and Adversarial Challenges in the Era of AI-Gene...

  121. [129]

    arXiv preprint arXiv:2511.00256 , year=

    NaturalVoices: A Large-Scale, Spontaneous and Emotional Podcast Dataset for Voice Conversion , author=. arXiv preprint arXiv:2511.00256 , year=

  122. [130]

    GenVC: Self-Supervised Zero-Shot Voice Conversion , booktitle =

    Zexin Cai and Henry Li Xinyuan and Ashi Garg and Leibny Paola Garc. GenVC: Self-Supervised Zero-Shot Voice Conversion , booktitle =. 2025 , url =. doi:10.1109/ASRU65441.2025.11434735 , timestamp =

    2025

  123. [131]

    arXiv preprint arXiv:2406.04494 , year=

    Towards naturalistic voice conversion: NaturalVoices dataset with an automatic processing pipeline , author=. arXiv preprint arXiv:2406.04494 , year=

    arXiv

  124. [132]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Dddm-vc: Decoupled denoising diffusion models with disentangled representation and prior mixup for verified robust voice conversion , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  125. [133]

    arXiv preprint arXiv:2507.13264 , year=

    Voxtral , author=. arXiv preprint arXiv:2507.13264 , year=

    arXiv

  126. [134]

    arXiv preprint arXiv:2509.17765 , year=

    Qwen3-omni technical report , author=. arXiv preprint arXiv:2509.17765 , year=

    arXiv

  127. [135]

    arXiv preprint arXiv:2601.15621 , year=

    Qwen3-TTS Technical Report , author=. arXiv preprint arXiv:2601.15621 , year=

    arXiv

  128. [136]

    5-Omni Technical Report , author=

    Qwen2. 5-Omni Technical Report , author=. arXiv preprint arXiv:2503.20215 , year=

    arXiv

  129. [137]

    arXiv preprint arXiv:2412.10117 , year=

    Cosyvoice 2: Scalable streaming speech synthesis with large language models , author=. arXiv preprint arXiv:2412.10117 , year=

    arXiv

  130. [138]

    arXiv preprint arXiv:2505.17589 , year=

    Cosyvoice 3: Towards in-the-wild speech generation via scaling-up and post-training , author=. arXiv preprint arXiv:2505.17589 , year=

    arXiv

  131. [139]

    TriAAN-VC: Triple Adaptive Attention Normalization for Any-to-Any Voice Conversion , year=

    Park, Hyun Joon and Woo Yang, Seok and Kim, Jin Sob and Shin, Wooseok and Han, Sung Won , booktitle=. TriAAN-VC: Triple Adaptive Attention Normalization for Any-to-Any Voice Conversion , year=

  132. [140]

    Using Joint Training Speaker Encoder With Consistency Loss to Achieve Cross-Lingual Voice Conversion and Expressive Voice Conversion , year=

    Guo, Houjian and Liu, Chaoran and Ishi, Carlos Toshinori and Ishiguro, Hiroshi , booktitle=. Using Joint Training Speaker Encoder With Consistency Loss to Achieve Cross-Lingual Voice Conversion and Expressive Voice Conversion , year=

  133. [141]

    arXiv preprint arXiv:2504.18425 , year=

    Kimi-audio technical report , author=. arXiv preprint arXiv:2504.18425 , year=

    arXiv

  134. [142]

    arXiv preprint arXiv:2408.01800 , year=

    Minicpm-v: A gpt-4v level mllm on your phone , author=. arXiv preprint arXiv:2408.01800 , year=

    arXiv

  135. [143]

    IEEE transactions on audio, speech, and language processing , volume=

    Analysis of emotionally salient aspects of fundamental frequency for emotion detection , author=. IEEE transactions on audio, speech, and language processing , volume=. 2009 , publisher=

    2009

  136. [144]

    ISCA & IEEE Workshop on Spontaneous Speech Processing and Recognition , pages=

    Experiments in emotional speech , author=. ISCA & IEEE Workshop on Spontaneous Speech Processing and Recognition , pages=

  137. [145]

    Speech communication , volume=

    Vocal communication of emotion: A review of research paradigms , author=. Speech communication , volume=. 2003 , publisher=

    2003

  138. [146]

    Vevo2: A Unified and Controllable Framework for Speech and Singing Voice Generation , year=

    Zhang, Xueyao and Zhang, Junan and Wang, Yuancheng and Wang, Chaoren and Chen, Yuanzhe and Jia, Dongya and Chen, Zhuo and Wu, Zhizheng , journal=. Vevo2: A Unified and Controllable Framework for Speech and Singing Voice Generation , year=

  139. [147]

    Emilia: An Extensive, Multilingual, and Diverse Speech Dataset For Large-Scale Speech Generation , year=

    He, Haorui and Shang, Zengqiang and Wang, Chaoren and Li, Xuyuan and Gu, Yicheng and Hua, Hua and Liu, Liwei and Yang, Chen and Li, Jiaqi and Shi, Peiyang and Wang, Yuancheng and Chen, Kai and Zhang, Pengyuan and Wu, Zhizheng , booktitle=. Emilia: An Extensive, Multilingual, a...

  140. [148]

    2025 , eprint=

    SingNet: Towards a Large-Scale, Diverse, and In-the-Wild Singing Voice Dataset , author=. 2025 , eprint=

    2025

  141. [149]

    Advancing Zero-shot Text-to-Speech Intelligibility across Diverse Domains via Preference Alignment

    Zhang, Xueyao and Wang, Yuancheng and Wang, Chaoren and Li, Ziniu and Chen, Zhuo and Wu, Zhizheng. Advancing Zero-shot Text-to-Speech Intelligibility across Diverse Domains via Preference Alignment. Proceedings of the 63rd Annual Meeting of the Association for Computational Li...

    2025 doi

  142. [150]

    Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

    M4Singer: A Multi-Style, Multi-Singer and Musical Score Provided Mandarin Singing Corpus , author=. Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

  143. [151]

    2025 , eprint=

    Qwen2.5 Technical Report , author=. 2025 , eprint=

    2025

  144. [152]

    arXiv preprint arXiv:2312.15185 , year=

    emotion2vec: Self-Supervised Pre-Training for Speech Emotion Representation , author=. arXiv preprint arXiv:2312.15185 , year=

    arXiv

  145. [153]

    doi:10.21437/Interspeech.2024-788 , issn =

    Ziyang Ma and Mingjie Chen and Hezhao Zhang and Zhisheng Zheng and Wenxi Chen and Xiquan Li and Jiaxin Ye and Xie Chen and Thomas Hain , year =. doi:10.21437/Interspeech.2024-788 , issn =

    2024 doi

  146. [154]

    2026 , eprint=

    Discovering and Causally Validating Emotion-Sensitive Neurons in Large Audio-Language Models , author=. 2026 , eprint=

    2026

  147. [155]

    2026 , eprint=

    Neuron-Level Emotion Control in Speech-Generative Large Audio-Language Models , author=. 2026 , eprint=

    2026

  148. [156]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    1972

  149. [157]

    Publications Manual , year = "1983", publisher =

    1983

  150. [158]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    1981

  151. [159]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  152. [160]

    Dan Gusfield , title =. 1997

    1997

  153. [161]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  154. [162]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.