Slide
Centre Interdisciplinaire
de Recherche et d’Innovation
en Cybersécurité et Société
de Recherche et d’Innovation
en Cybersécurité et Société
1.
Pimentel, A.; Zhu, Y.; Falk, T. H.
Partial Audio Deepfake Detection: Are We Really Detecting Synthetic Media or Just Dataset Content Biases? Proceedings Article
In: IEEE Conf. Artif. Intell., CAI, pp. 1846–1851, Institute of Electrical and Electronics Engineers Inc., 2026, ISBN: 979-833156039-3 (ISBN), (Journal Abbreviation: IEEE Conf. Artif. Intell., CAI).
Abstract | Links | BibTeX | Tags: Audio signal processing, Computational linguistics, Condition, Detection models, Information integrity, Language model, Large datasets, Learning models, Optimistics, Performance, Self-supervised learning, Speech communication, Speech models, Speech recognition, Speech signals, Synthetic media, Transcription
@inproceedings{pimentelPartialAudioDeepfake2026,
title = {Partial Audio Deepfake Detection: Are We Really Detecting Synthetic Media or Just Dataset Content Biases?},
author = {A. Pimentel and Y. Zhu and T. H. Falk},
url = {https://www.scopus.com/pages/publications/105042045476?origin=resultslist},
doi = {10.1109/CAI68641.2026.11536622},
isbn = {979-833156039-3 (ISBN)},
year = {2026},
date = {2026-01-01},
booktitle = {IEEE Conf. Artif. Intell., CAI},
pages = {1846–1851},
publisher = {Institute of Electrical and Electronics Engineers Inc.},
abstract = {In recent years, the generation of highly realistic audio deepfakes has raised significant concerns regarding privacy and information integrity. While most research has focused on fully bonafide or spoofed speech, partial deep-fakes, where only segments of an utterance are manipulated, remain less explored. Given the nature of the task, existing datasets are relying on large language models to manipulate bonafide speech signals into partial deepfakes by altering, deleting, or replacing segments with synthetic content. These manipulations may alter the semantics and sentiment of the generated content, creating biases that can be captured by deepfake detection models, leading to overly optimistic performance and poor generalizability. In this work, we investigate the presence of such biases in two popular partial deepfake datasets, namely AV-Deepfake1M and PartialEdit. We explore four self-supervised learning models, two relying only on text transcriptions (RoBERTa and RoBERTa-sentiment) and two relying on universal speech representations (WavLM Large and Wav2vec2-XLSR). Our experiments show that, while speech models achieve higher in-domain performance, they do not generalize to out-of-domain conditions. In turn, models trained on only transcribed speech can effectively distinguish manipulated content, achieving up to 97.8% AUC in-domain and nearly 70% AUC out-of-domain. These results suggest that linguistic patterns and dataset confounds may indeed be biasing partial deepfake detection models, leading to poor generalizability. © 2026 IEEE.},
note = {Journal Abbreviation: IEEE Conf. Artif. Intell., CAI},
keywords = {Audio signal processing, Computational linguistics, Condition, Detection models, Information integrity, Language model, Large datasets, Learning models, Optimistics, Performance, Self-supervised learning, Speech communication, Speech models, Speech recognition, Speech signals, Synthetic media, Transcription},
pubstate = {published},
tppubtype = {inproceedings}
}
In recent years, the generation of highly realistic audio deepfakes has raised significant concerns regarding privacy and information integrity. While most research has focused on fully bonafide or spoofed speech, partial deep-fakes, where only segments of an utterance are manipulated, remain less explored. Given the nature of the task, existing datasets are relying on large language models to manipulate bonafide speech signals into partial deepfakes by altering, deleting, or replacing segments with synthetic content. These manipulations may alter the semantics and sentiment of the generated content, creating biases that can be captured by deepfake detection models, leading to overly optimistic performance and poor generalizability. In this work, we investigate the presence of such biases in two popular partial deepfake datasets, namely AV-Deepfake1M and PartialEdit. We explore four self-supervised learning models, two relying only on text transcriptions (RoBERTa and RoBERTa-sentiment) and two relying on universal speech representations (WavLM Large and Wav2vec2-XLSR). Our experiments show that, while speech models achieve higher in-domain performance, they do not generalize to out-of-domain conditions. In turn, models trained on only transcribed speech can effectively distinguish manipulated content, achieving up to 97.8% AUC in-domain and nearly 70% AUC out-of-domain. These results suggest that linguistic patterns and dataset confounds may indeed be biasing partial deepfake detection models, leading to poor generalizability. © 2026 IEEE.
2.
Zhu, Y.; Davoust, A.; Falk, T. H.
DeepSick: Deceiving Voice-Based Diagnostic Models with Synthetic Multilingual Pathological Speech Signals Proceedings Article
In: Conf. Proc. IEEE Int. Conf. Syst. Man Cybern., pp. 69–74, Institute of Electrical and Electronics Engineers Inc., 2025, ISBN: 1062922X (ISSN); 979-833153358-8 (ISBN), (Journal Abbreviation: Conf. Proc. IEEE Int. Conf. Syst. Man Cybern.).
Abstract | Links | BibTeX | Tags: COVID-19, Detection models, Diagnosis, Diagnostic model, Diagnostic systems, Generative model, Health assessments, Pathological conditions, Pathological speech signals, Scalable solution, Speech communication, Speech recognition, Speech synthesis, State of the art, Voice model
@inproceedings{zhuDeepSickDeceivingVoiceBased2025,
title = {DeepSick: Deceiving Voice-Based Diagnostic Models with Synthetic Multilingual Pathological Speech Signals},
author = {Y. Zhu and A. Davoust and T. H. Falk},
url = {https://www.scopus.com/pages/publications/105033143787?origin=resultslist},
doi = {10.1109/SMC58881.2025.11343240},
isbn = {1062922X (ISSN); 979-833153358-8 (ISBN)},
year = {2025},
date = {2025-01-01},
booktitle = {Conf. Proc. IEEE Int. Conf. Syst. Man Cybern.},
pages = {69–74},
publisher = {Institute of Electrical and Electronics Engineers Inc.},
abstract = {Voice-based diagnostic systems offer a scalable solution for remote health assessment. However, recent advances in generative voice models may enable malicious manipulation of voice samples to simulate or conceal disease-related speech characteristics, which poses new risks to diagnostic systems. This paper investigates the vulnerability of diagnostic and detection models to such types of "deepfake"attacks. We show that it is possible to train a generative model to convert between healthy voices and pathological ones, which in turn, can successfully deceive existing diagnostic systems. Here, focus is placed on COVID-19 infection and respiratory abnormalities, but the method can be applied across different pathological conditions affecting vocal attributes. We also benchmark four state-of-the-art synthesized voice detection models on both real and generated pathological speech from three datasets. Our results show that current synthetic voice detectors, typically trained on healthy speech data, perform poorly on generated pathological samples. While fine-tuning with real pathological voices improves detection, a substantial performance gap remains. This work provides initial insights on an emerging threat to remote voice diagnostic systems that needs further work. © 2025 IEEE.},
note = {Journal Abbreviation: Conf. Proc. IEEE Int. Conf. Syst. Man Cybern.},
keywords = {COVID-19, Detection models, Diagnosis, Diagnostic model, Diagnostic systems, Generative model, Health assessments, Pathological conditions, Pathological speech signals, Scalable solution, Speech communication, Speech recognition, Speech synthesis, State of the art, Voice model},
pubstate = {published},
tppubtype = {inproceedings}
}
Voice-based diagnostic systems offer a scalable solution for remote health assessment. However, recent advances in generative voice models may enable malicious manipulation of voice samples to simulate or conceal disease-related speech characteristics, which poses new risks to diagnostic systems. This paper investigates the vulnerability of diagnostic and detection models to such types of "deepfake"attacks. We show that it is possible to train a generative model to convert between healthy voices and pathological ones, which in turn, can successfully deceive existing diagnostic systems. Here, focus is placed on COVID-19 infection and respiratory abnormalities, but the method can be applied across different pathological conditions affecting vocal attributes. We also benchmark four state-of-the-art synthesized voice detection models on both real and generated pathological speech from three datasets. Our results show that current synthetic voice detectors, typically trained on healthy speech data, perform poorly on generated pathological samples. While fine-tuning with real pathological voices improves detection, a substantial performance gap remains. This work provides initial insights on an emerging threat to remote voice diagnostic systems that needs further work. © 2025 IEEE.