Joint Fullband-Subband Modeling for High-Resolution SingFake Detection
Abstract
Rapid advances in singing voice synthesis have increased unauthorized imitation risks, creating an urgent need for better Singing Voice Deepfake (SingFake) Detection, also known as SVDD. Unlike speech, singing contains complex pitch, wide dynamic range, and timbral variations. Conventional 16 kHz-sampled detectors prove inadequate, as they discard vital high-frequency information. This study presents the first systematic analysis of high-resolution (44.1 kHz sampling rate) audio for SVDD. We propose a joint fullband-subband modeling framework: the fullband captures global context, while subband-specific experts isolate fine-grained synthesis artifacts unevenly distributed across the spectrum. Experiments on the WildSVDD dataset demonstrate that high-frequency subbands provide essential complementary cues. Our framework significantly outperforms 16 kHz-sampled models, proving that high-resolution audio and strategic subband integration are critical for robust in-the-wild detection.
The following analysis is generated by AI and may contain errors. It is separate from the authors’ abstract and is not peer review.
Research summary
This study presents a pioneering approach to high-resolution SingFake detection through a joint fullband-subband modeling framework, significantly enhancing the robustness of detection systems against sophisticated singing voice deepfakes. The innovative methodology and comprehensive experimental validation mark a substantial contribution to the field of audio deepfake detection.
Methodology
The proposed Sing-HiResNet framework introduces a novel joint fullband-subband modeling approach, effectively capturing both global context and localized artifacts in high-resolution audio. The methodology is well-structured, with clear phases for expert modeling and fusion strategies. The use of subband experts to isolate specific frequency ranges is a significant advancement over traditional fullband-only models, and the systematic evaluation of fusion strategies adds depth to the analysis. The approach is innovative and addresses a critical gap in the existing literature on SingFake detection.
Results & evaluation
The experiments conducted on the WildSVDD dataset demonstrate a comprehensive evaluation of the proposed framework, with a strong focus on performance metrics such as equal error rate (EER). The results indicate that the proposed model significantly outperforms existing 16 kHz models, showcasing the importance of high-resolution audio in detecting deepfakes. The comparative analysis against state-of-the-art systems further validates the effectiveness of the proposed methods.
Reproducibility
The paper provides sufficient details on the experimental setup, including dataset preparation, model architecture, and training procedures, which enhances reproducibility. However, the absence of publicly available code or a demo URL limits the ability for others to fully replicate the results.
Limitations
One limitation is the reliance on the WildSVDD dataset, which may not encompass the full diversity of real-world singing scenarios. Additionally, while the paper discusses the potential of high-frequency cues, the effectiveness of the model in extremely noisy environments or with varied vocal styles remains untested.
Broader impact
The implications of this work extend to various applications in audio forensics, copyright protection, and the broader field of deepfake detection. By improving SingFake detection, the research contributes to safeguarding artistic integrity and combating misinformation in media. The methodology could also inspire future research in audio signal processing and machine learning applications in other domains. This study presents a pioneering approach to high-resolution SingFake detection through a joint fullband-subband modeling framework, significantly enhancing the robustness of detection systems against sophisticated singing voice deepfakes. The innovative methodology and comprehensive experimental validation mark a substantial contribution to the field of audio deepfake detection.
Author of this paper? Send a correction.