Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2309.02232

FSD: An Initial Chinese Dataset for Fake Song Detection

Yuankun Xie·Jingjing Zhou·Xiaolin Lu·Zhenghao Jiang·Yuxin Yang·Haonan Cheng·Long Ye
View original source ↗Read source document ↗Export metadata ↓

Abstract

Singing voice synthesis and singing voice conversion have significantly advanced, revolutionizing musical experiences. However, the rise of "Deepfake Songs" generated by these technologies raises concerns about authenticity. Unlike Audio DeepFake Detection (ADD), the field of song deepfake detection lacks specialized datasets or methods for song authenticity verification. In this paper, we initially construct a Chinese Fake Song Detection (FSD) dataset to investigate the field of song deepfake detection. The fake songs in the FSD dataset are generated by five state-of-the-art singing voice synthesis and singing voice conversion methods. Our initial experiments on FSD revealed the ineffectiveness of existing speech-trained ADD models for the task of song deepFake detection. Thus, we employ the FSD dataset for the training of ADD models. We subsequently evaluate these models under two scenarios: one with the original songs and another with separated vocal tracks. Experiment results show that song-trained ADD models exhibit a 38.58% reduction in average equal error rate compared to speech-trained ADD models on the FSD test set.

Explore related subjects

cs.SDcs.AIeess.AS

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Yuankun Xie, Jingjing Zhou, Xiaolin Lu, Zhenghao Jiang, Yuxin Yang, Haonan Cheng, Long Ye. 2023-09-06. FSD: An Initial Chinese Dataset for Fake Song Detection. https://arxiv.org/abs/2309.02232

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Controllable Dysarthric Speech Synthesis with Patient-Specific Conditioning for Speaker-Diverse ASR Augmentation

Dysarthric speech recognition is limited by high speaker variability and scarce labeled data. Existing synthesis methods often couple speaker identity with dysarthric articulation, reducing control over generated speech. We propose a controllable dysarthric speech synthesis framework for ASR augmentation with separate prompt-derived timbre prefixes and learnable patient-specific pathology prefixes. Built on a pre-trained neural codec language model adapted using LoRA, the framework combines both prefixes through additive conditioning. A dual-classifier objective with gradient reversal and voice-conversion-based counterfactual augmentation promotes factor separation, allowing learned pathology conditions to be paired with different target speakers, including healthy speakers. Experiments on TORGO show that generated speech can partially replace real dysarthric training data and provides effective speaker-diverse augmentation when combined with real data. Objective ASR, perceptual, factor-separation, and phoneme-level analyses indicate preservation of target-speaker timbre and pathology-dependent patterns consistent with real dysarthric speech.

cs.SD↗
arXiv2026

How Contrastive Decoding Enhances Large Audio Language Models

While Contrastive Decoding (CD) has been proposed to enhance Large Audio Language Models (LALMs), it has not been evaluated at scale, and the underlying mechanisms driving its success remain unclear. This study systematically evaluates four distinct CD strategies across diverse LALM architectures. We identify Audio-Aware Decoding and Audio Contrastive Decoding as the most effective methods. However, their impact varies significantly across models. To explain this variability, we profile the baseline error composition of each model and measure how readily contrastive decoding corrects each error type. Our analysis demonstrates that CD reliably rectifies errors in which models falsely claim an absence of audio or resort to uncertainty-driven guessing, but is relatively poor at correcting flawed reasoning or confident misassertions. Crucially, CD's benefit closely tracks the composition of a model's baseline error profile: when errors caused by audio ignorance or uncertainty-driven guessing constitute only a small fraction of a model's errors, gains are marginal or even negative. A token-level analysis reveals the underlying mechanism: when the amateur's output is dominated by hesitation markers, CD's suppression naturally targets uncertainty-driven errors while having limited effect on confident misassertions.

cs.SD↗
arXiv2026

Multimodal Dataset Normalization and Perceptual Validation for Music-Taste Correspondences

Music and food traditions are both intangible cultural heritage, and the links between them, how a sound can make a taste seem sweeter or more bitter, are increasingly used in museum, exhibition and gastronomic-tourism settings. Modelling those links computationally runs into a data bottleneck familiar across cultural heritage computing: expert annotation is slow and costly, so the annotated collections that result are small. The usual remedy is to enlarge a collection automatically, labelling it with a model trained on the small annotated one. Such synthetic labels are rarely checked, either against the original annotations or against people. We provide both checks. Experiment 1 asks whether the audio-flavour patterns found in an experimental soundtrack collection (257 tracks annotated by listeners) survive when the collection is scaled to 49,300 30-second segments from the Free Music Archive (FMA) labelled by a fine-tuned Audio Spectrogram Transformer. Experiment 2 asks whether flavour profiles computed from food chemistry, for 20 dishes drawn largely from Italian culinary tradition, match what listeners actually hear (49 participants, online). Feature-flavour patterns carry over for every taste dimension (rho=0.38-0.72, all p<0.001), and sweetness still carries over when every spectral feature is removed, so the agreement is not an artefact of how the labelling model represents audio. Listener ratings match the computed profiles far beyond chance (permutation p<0.001; Mantel r=0.45; Procrustes m^2=0.49), and the result holds when participants reporting hearing or taste impairments are excluded. We release the harmonized datasets and all code.

cs.SD↗