Detail výsledku
LlamaPartialSpoof: An LLM-Driven Fake Speech Dataset Simulating Disinformation Generation
Li Haoyang
Zhang Lin, Ph.D.
Lee Kong Aik
Chng Eng Siong
Previous fake speech datasets were constructed from a defender's perspective to develop countermeasure (CM) systems without considering diverse motivations of attackers. To better align with real-life scenarios, we created LlamaPartialSpoof, a 130-hour dataset that contains both fully and partially fake speech, using a large language model (LLM) and voice cloning technologies to evaluate the robustness of CMs. By examining valuable information for both attackers and defenders, we identify several key vulnerabilities in current CM systems, which can be exploited to enhance attack success rates, including biases toward certain text-to-speech models or concatenation methods. Our experimental results indicate that the current fake speech detection system struggle to generalize to unseen scenarios, achieving a best performance of 24.49% equal error rate.
dataset | deepfake | fake speech detection | large language model | voice cloning
@inproceedings{BUT199992,
author="{} and {} and Lin {Zhang} and {} and {}",
title="LlamaPartialSpoof: An LLM-Driven Fake Speech Dataset Simulating Disinformation Generation",
booktitle="Proceedings - ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing",
year="2025",
pages="5",
publisher="Institute of Electrical and Electronics Engineers Inc.",
address="Hyderabad, Indická republika",
doi="10.1109/ICASSP49660.2025.10888070",
isbn="979-8-3503-6874-1",
url="https://ieeexplore.ieee.org/document/10888070"
}