Recently, binaural audio synthesis (BAS) has emerged as a promising research field for its applications in augmented and virtual realities. Binaural audio helps us to orient ourselves and establish immersion by providing the brain with interaural time differences reflecting spatial information. However, existing methods are limited in terms of phase estimation, which is crucial for spatial hearing. In this paper, we propose the DopplerBAS method to explicitly address the Doppler effect of the moving sound source. Specifically, we calculate the radial relative velocity of the moving speaker in spherical coordinates, which further guides the synthesis of binaural audio. This simple method neither introduces any additional hyper-parameters nor modifies the loss functions, and is plug-and-play: it scales well to different types of backbones. DopplerBAS distinctly improves WarpNet and BinauralGrad in the phase error metric and reaches a new state-of-the-art: 0.780 (vs. the current state-of-the-art 0.807). Experiments and ablation studies demonstrate the effectiveness of our method.
Out-of-distribution (OOD) detection is an important task to ensure the reliability and safety of deep learning and the discriminator models outperform others for now. However, the feature extraction of the discriminator models must compress the data and lose certain information, leaving room for bad cases and malicious attacks. In this paper, we provide a new assumption that the discriminator models are more sensitive to some subareas of the input space and such perceptron bias causes bad cases and overconfidence areas. Under this assumption, we design new detection methods and indicator scores. For detection methods, we introduce diffusion models (DMs) into OOD detection. We find that the diffusion denoising process (DDP) of DMs also functions as a novel form of asymmetric interpolation, which is suitable to enhance the input and reduce the overconfidence areas. For indicator scores, we find that the features of the discriminator models of OOD inputs occur sharp changes under DDP and use the norm of this dynamic change as our indicator scores. Therefore, we develop a new framework to combine the discriminator and generation models to do OOD detection under our new assumption. The discriminator models provide proper detection spaces and the generation models reduce the overconfidence problem. According to our experiments on CIFAR10 and CIFAR100, our methods get competitive results with state-of-the-art methods. Our implementation is available at https://github.com/luping-liu/DiffOOD.
Video to sound generation aims to generate realistic and natural sound given a video input. However, previous video-to-sound generation methods can only generate a random or average timbre without any controls or specializations of the generated sound timbre, leading to the problem that people cannot obtain the desired timbre under these methods sometimes. In this paper, we pose the task of generating sound with a specific timbre given a video input and a reference audio sample. To solve this task, we disentangle each target sound audio into three components: temporal information, acoustic information, and background information. We first use three encoders to encode these components respectively: 1) a temporal encoder to encode temporal information, which is fed with video frames since the input video shares the same temporal information as the original audio; 2) an acoustic encoder to encode timbre information, which takes the original audio as input and discards its temporal information by a temporal-corrupting operation; and 3) a background encoder to encode the residual or background sound, which uses the background part of the original audio as input. To make the generated result achieve better quality and temporal alignment, we also adopt a mel discriminator and a temporal discriminator for the adversarial training. Our experimental results on the VAS dataset demonstrate that our method can generate high-quality audio samples with good synchronization with events in video and high timbre similarity with the reference audio.
Multi-modal video question answering aims to predict correct answer and localize the temporal boundary relevant to the question. The temporal annotations of questions improve QA performance and interpretability of recent works, but they are usually empirical and costly. To avoid the temporal annotations, we devise a weakly supervised question grounding (WSQG) setting, where only QA annotations are used and the relevant temporal boundaries are generated according to the temporal attention scores. To substitute the temporal annotations, we transform the correspondence between frames and subtitles to Frame-Subtitle (FS) self-supervision, which helps to optimize the temporal attention scores and hence improve the video-language understanding in VideoQA model. The extensive experiments on TVQA and TVQA+ datasets demonstrate that the proposed WSQG strategy gets comparable performance on question grounding, and the FS self-supervision helps improve the question answering and grounding performance on both QA-supervision only and full-supervision settings.
The task of argument mining aims to detect all possible argumentative components and identify their relationships automatically. As a thriving field in natural language processing, there has been a large amount of corpus for academic study and application development in argument mining. However, the research in this area is still constrained by the inherent limitations of existing datasets. Specifically, all the publicly available datasets are relatively small in scale, and few of them provide information from other modalities to facilitate the learning process. Moreover, the statements and expressions in these corpora are usually in a compact form, which means non-adjacent clauses or text segments will always be regarded as multiple individual components, thus restricting the generalization ability of models. To this end, we collect and contribute a novel dataset AntCritic to serve as a helpful complement to this area, which consists of about 10k free-form and visually-rich financial comments and supports both argument component detection and argument relation prediction tasks. Besides, in order to cope with the challenges and difficulties brought by scenario expansion and problem setting modification, we thoroughly explore the fine-grained relation prediction and structure reconstruction scheme for free-form documents and discuss the encoding mechanism for visual styles and layouts. And based on these analyses, we design two simple but effective model architectures and conduct various experiments on this dataset to provide benchmark performances as a reference and verify the practicability of our proposed architecture.
Micro-video recommender systems suffer from the ubiquitous noises in users' behaviors, which might render the learned user representation indiscriminating, and lead to trivial recommendations (e.g., popular items) or even weird ones that are far beyond users' interests. Contrastive learning is an emergent technique for learning discriminating representations with random data augmentations. However, due to neglecting the noises in user behaviors and treating all augmented samples equally, the existing contrastive learning framework is insufficient for learning discriminating user representations in recommendation. To bridge this research gap, we propose the Contrast over Contrastive Learning framework for training recommender models, named CCL4Rec, which models the nuances of different augmented views by further contrasting augmented positives/negatives with adaptive pulling/pushing strengths, i.e., the contrast over (vanilla) contrastive learning. To accommodate these contrasts, we devise the hardness-aware augmentations that track the importance of behaviors being replaced in the query user and the relatedness of substitutes, and thus determining the quality of augmented positives/negatives. The hardness-aware augmentation also permits controllable contrastive learning, leading to performance gains and robust training. In this way, CCL4Rec captures the nuances of historical behaviors for a given user, which explicitly shields off the learned user representation from the effects of noisy behaviors. We conduct extensive experiments on two micro-video recommendation benchmarks, which demonstrate that CCL4Rec with far less model parameters could achieve comparable performance to existing state-of-the-art method, and improve the training/inference speed by several orders of magnitude.
Effectively representing users lie at the core of modern recommender systems. Since users' interests naturally exhibit multiple aspects, it is of increasing interest to develop multi-interest frameworks for recommendation, rather than represent each user with an overall embedding. Despite their effectiveness, existing methods solely exploit the encoder (the forward flow) to represent multiple aspects of interests. However, without explicit regularization, the interest embeddings may not be distinct from each other nor semantically reflect representative historical items. Towards this end, we propose the Re4 framework, which leverages the backward flow to reexamine each interest embedding. Specifically, Re4 encapsulates three backward flows, i.e., 1) Re-contrast, which drives each interest embedding to be distinct from other interests using contrastive learning; 2) Re-attend, which ensures the interest-item correlation estimation in the forward flow to be consistent with the criterion used in final recommendation; and 3) Re-construct, which ensures that each interest embedding can semantically reflect the information of representative items that relate to the corresponding interest. We demonstrate the novel forward-backward multi-interest paradigm on ComiRec, and perform extensive experiments on three real-world datasets. Empirical studies validate that Re4 helps to learn learning distinct and effective multi-interest representations.
Denoising diffusion probabilistic models (DDPMs) have recently achieved leading performances in many generative tasks. However, the inherited iterative sampling process costs hinder their applications to text-to-speech deployment. Through the preliminary study on diffusion model parameterization, we find that previous gradient-based TTS models require hundreds or thousands of iterations to guarantee high sample quality, which poses a challenge for accelerating sampling. In this work, we propose ProDiff, on progressive fast diffusion model for high-quality text-to-speech. Unlike previous work estimating the gradient for data density, ProDiff parameterizes the denoising model by directly predicting clean data to avoid distinct quality degradation in accelerating sampling. To tackle the model convergence challenge with decreased diffusion iterations, ProDiff reduces the data variance in the target site via knowledge distillation. Specifically, the denoising model uses the generated mel-spectrogram from an N-step DDIM teacher as the training target and distills the behavior into a new model with N/2 steps. As such, it allows the TTS model to make sharp predictions and further reduces the sampling time by orders of magnitude. Our evaluation demonstrates that ProDiff needs only 2 iterations to synthesize high-fidelity mel-spectrograms, while it maintains sample quality and diversity competitive with state-of-the-art models using hundreds of steps. ProDiff enables a sampling speed of 24x faster than real-time on a single NVIDIA 2080Ti GPU, making diffusion models practically applicable to text-to-speech synthesis deployment for the first time. Our extensive ablation studies demonstrate that each design in ProDiff is effective, and we further show that ProDiff can be easily extended to the multi-speaker setting. Audio samples are available at \url{https://ProDiff.github.io/.}
Unconstrained lip-to-speech synthesis aims to generate corresponding speeches from silent videos of talking faces with no restriction on head poses or vocabulary. Current works mainly use sequence-to-sequence models to solve this problem, either in an autoregressive architecture or a flow-based non-autoregressive architecture. However, these models suffer from several drawbacks: 1) Instead of directly generating audios, they use a two-stage pipeline that first generates mel-spectrograms and then reconstructs audios from the spectrograms. This causes cumbersome deployment and degradation of speech quality due to error propagation; 2) The audio reconstruction algorithm used by these models limits the inference speed and audio quality, while neural vocoders are not available for these models since their output spectrograms are not accurate enough; 3) The autoregressive model suffers from high inference latency, while the flow-based model has high memory occupancy: neither of them is efficient enough in both time and memory usage. To tackle these problems, we propose FastLTS, a non-autoregressive end-to-end model which can directly synthesize high-quality speech audios from unconstrained talking videos with low latency, and has a relatively small model size. Besides, different from the widely used 3D-CNN visual frontend for lip movement encoding, we for the first time propose a transformer-based visual frontend for this task. Experiments show that our model achieves $19.76\times$ speedup for audio waveform generation compared with the current autoregressive model on input sequences of 3 seconds, and obtains superior audio quality.