Contrastive cross-modal models such as CLIP and CLAP aid various vision-language (VL) and audio-language (AL) tasks. However, there has been limited investigation of and improvement in their language encoder, which is the central component of encoding natural language descriptions of image/audio into vector representations. We extensively evaluate how unsupervised and supervised sentence embedding training affect language encoder quality and cross-modal task performance. In VL pretraining, we found that sentence embedding training language encoder quality and aids in cross-modal tasks, improving contrastive VL models such as CyCLIP. In contrast, AL pretraining benefits less from sentence embedding training, which may result from the limited amount of pretraining data. We analyze the representation spaces to understand the strengths of sentence embedding training, and find that it improves text-space uniformity, at the cost of decreased cross-modal alignment.
Diffusion-based speech enhancement (SE) has been investigated recently, but its decoding is very time-consuming. One solution is to initialize the decoding process with the enhanced feature estimated by a predictive SE system. However, this two-stage method ignores the complementarity between predictive and diffusion SE. In this paper, we propose a unified system that integrates these two SE modules. The system encodes both generative and predictive information, and then applies both generative and predictive decoders, whose outputs are fused. Specifically, the two SE modules are fused in the first and final diffusion steps: the first step fusion initializes the diffusion process with the predictive SE for improving the convergence, and the final step fusion combines the two complementary SE outputs to improve the SE performance. Experiments on the Voice-Bank dataset show that the diffusion score estimation can benefit from the predictive information and speed up the decoding.
Audio classification and restoration are among major downstream tasks in audio signal processing. However, restoration derives less of a benefit from pretrained models compared to the overwhelming success of pretrained models in classification tasks. Due to such unbalanced benefits, there has been rising interest in how to improve the performance of pretrained models for restoration tasks such as speech enhancement (SE). Previous works have shown that the features extracted by pretrained audio encoders are effective for SE tasks, but these speech-specific encoder-only models usually require extra decoders to become compatible with SE tasks, and involve complicated pretraining procedures or complex data augmentation. Therefore, in pursuit of a universal audio model, the audio masked autoencoder (MAE) whose backbone is the autoencoder of Vision Transformers (ViT-AE), is extended from audio classification toward restoration tasks in this paper. ViT-AE naturally learns mel-to-mel mapping that is compatible with restoration tasks during pretraining. Among many restoration tasks, SE is chosen due to its well-established evaluation metrics and test data. We propose variations of ViT-AE to improve the SE performance, where the mel-to-mel variations yield high scores for non-intrusive metrics and the STFT-oriented variation is effective at standard intrusive metrics such as PESQ. Different variations can be used in accordance with the scenarios. Comprehensive evaluations and ablation studies show that MAE pretraining is also beneficial to SE tasks and help the ViT-AE to better generalize to out-of-domain distortions. We further found that large-scale noisy data of general audio sources, rather than clean speech, is sufficiently effective for pretraining.
Although music is typically multi-label, many works have studied hierarchical music tagging with simplified settings such as single-label data. Moreover, there lacks a framework to describe various joint training methods under the multi-label setting. In order to discuss the above topics, we introduce hierarchical multi-label music instrument classification task. The task provides a realistic setting where multi-instrument real music data is assumed. Various hierarchical methods that jointly train a DNN are summarized and explored in the context of the fusion of deep learning and conventional techniques. For the effective joint training in the multi-label setting, we propose two methods to model the connection between fine- and coarse-level tags, where one uses rule-based grouped max-pooling, the other one uses the attention mechanism obtained in a data-driven manner. Our evaluation reveals that the proposed methods have advantages over the method without joint training. In addition, the decision procedure within the proposed methods can be interpreted by visualizing attention maps or referring to fixed rules.
Geosystems are geological formations altered by humans activities such as fossil energy exploration, waste disposal, geologic carbon sequestration, and renewable energy generation. Geosystems also represent a critical link in the global water-energy nexus, providing both the source and buffering mechanisms for enabling societal adaptation to climate variability and change. The responsible use and exploration of geosystems are thus critical to the geosystem governance, which in turn depends on the efficient monitoring, risk assessment, and decision support tools for practical implementation. Fast advances in machine learning (ML) algorithms and novel sensing technologies in recent years have presented new opportunities for the subsurface research community to improve the efficacy and transparency of geosystem governance. Although recent studies have shown the great promise of scientific ML (SciML) models, questions remain on how to best leverage ML in the management of geosystems, which are typified by multiscality, high-dimensionality, and data resolution inhomogeneity. This survey will provide a systematic review of the recent development and applications of domain-aware SciML in geosystem researches, with an emphasis on how the accuracy, interpretability, scalability, defensibility, and generalization skill of ML approaches can be improved to better serve the geoscientific community.
Global hydrological and land surface models are increasingly used for tracking terrestrial total water storage (TWS) dynamics, but the utility of existing models is hampered by conceptual and/or data uncertainties related to various underrepresented and unrepresented processes, such as groundwater storage. The gravity recovery and climate experiment (GRACE) satellite mission provided a valuable independent data source for tracking TWS at regional and continental scales. Strong interests exist in fusing GRACE data into global hydrological models to improve their predictive performance. Here we develop and apply deep convolutional neural network (CNN) models to learn the spatiotemporal patterns of mismatch between TWS anomalies (TWSA) derived from GRACE and those simulated by NOAH, a widely used land surface model. Once trained, our CNN models can be used to correct the NOAH simulated TWSA without requiring GRACE data, potentially filling the data gap between GRACE and its follow-on mission, GRACE-FO. Our methodology is demonstrated over India, which has experienced significant groundwater depletion in recent decades that is nevertheless not being captured by the NOAH model. Results show that the CNN models significantly improve the match with GRACE TWSA, achieving a country-average correlation coefficient of 0.94 and Nash-Sutcliff efficient of 0.87, or 14\% and 52\% improvement respectively over the original NOAH TWSA. At the local scale, the learned mismatch pattern correlates well with the observed in situ groundwater storage anomaly data for most parts of India, suggesting that deep learning models effectively compensate for the missing groundwater component in NOAH for this study region.