The development of speech synthesis technology has increased the attention toward the threat of spoofed speech. Although various high-performance spoofing countermeasures have been proposed in recent years, a particular scenario is overlooked: partially spoofed audio, where spoofed utterances may contain both spoofed and bona fide segments. Currently, the research on partially spoofed speech detection is lacking. The existing methods either train with partially spoofed speech at utterance level, resulting in gradient conflicting at the segment level, or directly train with segment level data, which requires segment labels that are difficult to obtain in practice. In this study, to better detect partially spoofed speech when only utterance labels are available, we formulate partially spoofed speech detection into a multiple instance learning (MIL) problem. The typical MIL uses a pooling layer to fuse patch scores as a whole, and we propose a hybrid MIL (H-MIL) framework based on max and log-sum-exp pooling methods, which can learn better segment representations to improve partially spoofed speech detection performance. Theoretical and experimental verification shows that H-MIL can effectively relieve the gradient conflicting and gradient vanishing problems. In addition, we analyze the local correlations between segments and introduce a local self-attention mechanism to enhance segment features, which further promotes the detection performance. In our experiments, we provide not only detection results at the segment and utterance levels but also some detailed visualization analysis, including the effect of spoof ratio and cross-dataset detection. The experimental results demonstrate the effective detection performance of our method at both the utterance and segment levels, especially when dealing with low spoof ratio attacks. The results confirm that our approach can better deal with partially spoofed speech detection than previous methods.

Highly deceptive deepfake technologies have caused much controversy, e.g., artificial intelligence-based software can automatically generate nude photos and deepfake images of anyone. This brings considerable threats to both individuals and society. In addition to video and image forgery, audio forgery poses many hazards but lacks sufficient attention. Furthermore, existing works have only focused on voice spoof detection, neglecting the identification of spoof algorithms. It is of great value to recognize the algorithm for synthesizing spoofing voices in traceability. This study presents a system combining voice spoof detection and algorithm recognition. In contrast, the generalizability of the spoof detection model is discussed from the perspective of embedding space and decision boundaries to face the voice spoofing attacks generated by spoof algorithms that are not available in the training set. This study presents a method for voice spoof algorithms recognition based on incremental learning, taking into account data flow scenarios where new spoof algorithms keep appearing in reality. Our experimental results on the LA dataset of ASVspoof show that our system can improve the generalization of spoof detection and identify new voice spoof algorithms without catastrophic forgetting.

Fine-grained visual categorization aims at differentiating subcategories such as different species of birds, models of cars and variants of aircraft. It often suffers from small inter-variance and large intra-variance. To keep dissimilar images far apart and preserve large intra-variance simultaneously, we propose an adaptive triplet model. At first, images are batched as triplets and input to a general convolutional network, which extracts convolutional image features. Then, we combine adaptive triplet loss and classification loss for multi-task training. Adaptive triplet loss pulls same-class embeddings together and pushes examples from different subcategories apart. It allocates different weights to hard and easy examples in an adaptive way in the training process. Unlike previous hard mining mechanisms that discard all nonhard triplets, it can benefit from all possible informative examples. Moreover, a second-order distance function is put forward to capture local pairwise interactions of embeddings, which is more discriminative in distance measure. Classification loss is used to provide more direct supervision for training embeddings with category specific concepts. Further, it makes the prediction of category more convenient and more efficient in testing. Experiments demonstrate state-of-the-art results on three popular fine-grained datasets, including CUB-200-2011, Stanford Cars and FGVC-Aircraft. In addition, our network structure is relatively simple compared with previous methods, which often suffer from multiple sub-networks and complex training mechanisms. It is also applicable for most up-to-date backbone networks, while others might be restricted to specific convolutional networks.

Due to the rise of many fields such as e-commerce platforms, a large number of stream data has emerged. The incomplete labeling problem and concept drift problem of these data pose a huge challenge to the existing stream data classification methods. In this respect, a dynamic stream data classification algorithm is proposed for the stream data. For the incomplete labeling problem, this method introduces randomization and iterative strategy based on the very fast decision tree VFDT algorithm to design an iterative integration algorithm, and the algorithm uses the previous model classification result as the next model input and implements the voting mechanism for new data classification. At the same time, the window mechanism is used to store data and calculate the data distribution characteristics in the window, then, combined with the calculated result and the predicted amount of data to adjust the size of the sliding window. Experiments show the superiority of the algorithm in classification accuracy. The aim of the study is to compare different algorithms to evaluate whether classification model adapts to the current data environment.