The rapid advancement of audio generation technologies has escalated the risks of malicious deepfake audio across speech, sound, singing voice, and music, threatening multimedia security and trust. While existing countermeasures (CMs) perform well in single-type audio deepfake detection (ADD), their performance declines in cross-type scenarios. This paper is dedicated to studying the all-type ADD task. We are the first to comprehensively establish an all-type ADD benchmark to evaluate current CMs, incorporating cross-type deepfake detection across speech, sound, singing voice, and music. Then, we introduce the prompt tuning self-supervised learning (PT-SSL) training paradigm, which optimizes SSL front-end by learning specialized prompt tokens for ADD, requiring 458× fewer trainable parameters than fine-tuning (FT). Considering the auditory perception of different audio types, we propose the wavelet prompt tuning (WPT)-SSL method to capture type-invariant auditory deepfake information from the frequency domain without requiring additional training parameters, thereby enhancing performance over FT in the all-type ADD task. To achieve an universally CM, we utilize all types of deepfake audio for co-training. Experimental results demonstrate that WPT-XLSR-AASIST achieved the best performance, with an average EER of 3.58% across all evaluation sets.

In recent years, speech diffusion models have advanced rapidly. Alongside the widely used U-Net architecture, transformer-based models such as the Diffusion Transformer (DiT) have also gained attention. However, current DiT speech models treat Mel spectrograms as general images, which overlooks the specific acoustic properties of speech. To address these limitations, we propose a method called Directional Patch Interaction for Text-to-Speech (DPI-TTS), which builds on DiT and achieves fast training without compromising accuracy. Notably, DPI-TTS employs a low-to-high frequency, frame-by-frame progressive inference approach that aligns more closely with acoustic properties, enhancing the naturalness of the generated speech. Additionally, we introduce a fine-grained style temporal modeling method that further improves speaker style similarity. Experimental results demonstrate that our method increases the training speed by nearly 2 times and significantly outperforms the baseline models.

Speech synthesis technology has posed a serious threat to speaker verification systems. Currently, the most effective fake audio detection methods utilize pretrained models, and integrating features from various layers of pretrained model further enhances detection performance. However, most of the previously proposed fusion methods require fine-tuning the pretrained models, resulting in excessively long training times and hindering model iteration when facing new speech synthesis technology. To address this issue, this paper proposes a feature fusion method based on the Mixture of Experts, which extracts and integrates features relevant to fake audio detection from layer features, guided by a gating network based on the last layer feature, while freezing the pretrained model. Experiments conducted on the ASVspoof2019 and ASVspoof2021 datasets demonstrate that the proposed method achieves competitive performance compared to those requiring fine-tuning.

Partially spoofed audio detection is a challenging task, lying in the need to accurately locate the authenticity of audio at the frame level. To address this issue, we propose a fine-grained partially spoofed audio detection method, namely Temporal Deepfake Location (TDL), which can effectively capture information of both features and locations. Specifically, our approach involves two novel parts: embedding similarity module and temporal convolution operation. To enhance the identification between the real and fake features, the embedding similarity module is designed to generate an embedding space that can separate the real frames from fake frames. To effectively concentrate on the position information, temporal convolution operation is proposed to calculate the frame-specific similarities among neighboring frames, and dynamically select informative neighbors to convolution. Extensive experiments show that our method outperform baseline models in ASVspoof2019 Partial Spoof dataset and demonstrate superior performance even in the cross-dataset scenario.

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.

In this paper, we propose an Aggregation and Separation Domain Generalization (ASDG) method for Audio DeepFake Detection (ADD). Fake speech generated from different methods exhibits varied amplitude and frequency distributions rather than genuine speech. In addition, the spoofing attacks in training sets may not keep pace with the evolving diversity of real-world deepfake distributions. In light of this, we attempt to learn an ideal feature space that can aggregate real speech and separate fake speech to achieve better generalizability in the detection of unseen target domains. Specifically, we first propose a feature generator based on Lightweight Convolutional Neural Networks (LCNN), which is employed for generating a feature space and categorizing the feature into real and fake. Meanwhile, single-side domain adversarial learning is leveraged to make only the real speech from different domains indistinguishable, which enables the distribution of real speech to be aggregated in the feature space. Furthermore, a triplet loss is adopted to separate the distribution of fake speech while aggregating the distribution of real speech. Finally, in order to test the generalizability of the model, we train it with three different English datasets and evaluate in harsh conditions: cross-language and noisy datasets. The extensive experiments show that ASDG outperforms the baseline models in cross-domain tasks and decreases Equal Error Rate (EER) by up to 39.24% when compared to that of RawNet2. It is proved that the proposed Aggregation and Separation Domain Generalization method can be an effective strategy to improve the model generalizability.

In this paper, we propose a novel prosody disentangle method for prosodic Text-to-Speech (TTS) model, which introduces the vector quantization (VQ) method to the auxiliary prosody encoder to obtain the decomposed prosody representations in an unsupervised manner. Rely on its advantages, the speaking styles, such as pitch, speaking velocity, local pitch variance, etc., are decomposed automatically into the latent quantize vectors. We also investigate the internal mechanism of VQ disentangle process by means of a latent variables counter and find that higher value dimensions usually represent prosody information. Experiments show that our model can control the speaking styles of synthesis results by directly manipulating the latent variables. The objective and subjective evaluations illustrated that our model outperforms the popular models.