How Well Do Current Speech Deepfake Detection Methods Generalize to the Real World?

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Li, Daixian, Xue, Jun, Ren, Yanzhen, Yi, Zhuolin, Huang, Yihuan, Feng, Guanxiang, Chai, Yi
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912947393003520
author Li, Daixian
Xue, Jun
Ren, Yanzhen
Yi, Zhuolin
Huang, Yihuan
Feng, Guanxiang
Chai, Yi
author_facet Li, Daixian
Xue, Jun
Ren, Yanzhen
Yi, Zhuolin
Huang, Yihuan
Feng, Guanxiang
Chai, Yi
contents Recent advances in speech synthesis and voice conversion have greatly improved the naturalness and authenticity of generated audio. Meanwhile, evolving encoding, compression, and transmission mechanisms on social media platforms further obscure deepfake artifacts. These factors complicate reliable detection in real-world environments, underscoring the need for representative evaluation benchmarks. To this end, we introduce ML-ITW (Multilingual In-The-Wild), a multilingual dataset covering 14 languages, seven major platforms, and 180 public figures, totaling 28.39 hours of audio. We evaluate three detection paradigms: end-to-end neural models, self-supervised feature-based (SSL) methods, and audio large language models (Audio LLMs). Experimental results reveal significant performance degradation across diverse languages and real-world acoustic conditions, highlighting the limited generalization ability of existing detectors in practical scenarios. The ML-ITW dataset is publicly available.
format Preprint
id arxiv_https___arxiv_org_abs_2603_05852
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle How Well Do Current Speech Deepfake Detection Methods Generalize to the Real World?
Li, Daixian
Xue, Jun
Ren, Yanzhen
Yi, Zhuolin
Huang, Yihuan
Feng, Guanxiang
Chai, Yi
Sound
Recent advances in speech synthesis and voice conversion have greatly improved the naturalness and authenticity of generated audio. Meanwhile, evolving encoding, compression, and transmission mechanisms on social media platforms further obscure deepfake artifacts. These factors complicate reliable detection in real-world environments, underscoring the need for representative evaluation benchmarks. To this end, we introduce ML-ITW (Multilingual In-The-Wild), a multilingual dataset covering 14 languages, seven major platforms, and 180 public figures, totaling 28.39 hours of audio. We evaluate three detection paradigms: end-to-end neural models, self-supervised feature-based (SSL) methods, and audio large language models (Audio LLMs). Experimental results reveal significant performance degradation across diverse languages and real-world acoustic conditions, highlighting the limited generalization ability of existing detectors in practical scenarios. The ML-ITW dataset is publicly available.
title How Well Do Current Speech Deepfake Detection Methods Generalize to the Real World?
topic Sound
url https://arxiv.org/abs/2603.05852