We introduce AnyEnhance, a unified generative model for voice enhancement that processes both speech and singing voices. Based on a masked generative model, AnyEnhance is capable of handling both speech and singing voices, supporting a wide range of enhancement tasks including denoising, dereverberation, declipping, super-resolution, and target speaker extraction, all simultaneously and without fine-tuning. AnyEnhance introduces a prompt-guidance mechanism for in-context learning, which allows the model to natively accept a reference speaker’s timbre. In this way, it could boost enhancement performance when a reference audio is available and enable the target speaker extraction task without altering the underlying architecture. Moreover, we also introduce a self-critic mechanism into the generative process for masked generative models, yielding higher-quality outputs through iterative self-assessment and refinement. Extensive experiments on various enhancement tasks demonstrate AnyEnhance outperforms existing methods in terms of both objective metrics and subjective listening tests.
13 March 2018 in ACM Transactions on Information Systems
Many social network applications depend on robust representations of spatio-temporal data. In this work, we present an embedding model based on feed-forward neural networks which transforms social media check-ins into dense feature vectors encoding geographic, temporal, and functional aspects for modeling places, neighborhoods, and users. We employ the embedding model in a variety of applications including location recommendation , urban functional zone study , and crime prediction . For location recommendation , we propose a Spatio-Temporal Embedding Similarity algorithm (STES) based on the embedding model. In a range of experiments on real life data collected from Foursquare, we demonstrate our model’s effectiveness at characterizing places and people and its applicability in aforementioned problem domains. Finally, we select eight major cities around the globe and verify the robustness and generality of our model by porting pre-trained models from one city to another, thereby alleviating the need for costly local training.
For next place prediction, machine learning methods which incorporate contextual data are frequently used. However, previous studies often do not allow deriving generalizable methodological recommendations, since they use different datasets, methods for discretizing space, scales of prediction, prediction algorithms, and context data, and therefore lack comparability. Additionally, the cold start problem for new users is an issue. In this study, we predict next places based on one trajectory dataset but with systematically varying prediction algorithms, methods for space discretization, scales of prediction (based on a novel hierarchical approach), and incorporated context data. This allows to evaluate the relative influence of these factors on the overall prediction accuracy. Moreover, in order to tackle the cold start problem prevalent in recommender and prediction systems, we test the effect of training the predictor on all users instead of each individual one. We find that the prediction accuracy shows a varying dependency on the method of space discretization and the incorporated contextual factors at different spatial scales. Moreover, our user-independent approach reaches a prediction accuracy of around 75%, and is therefore an alternative to existing user-specific models. This research provides valuable insights into the individual and combinatory effects of model parameters and algorithms on the next place prediction accuracy. The results presented in this paper can be used to determine the influence of various contextual factors and to help researchers building more accurate prediction models. It is also a starting point for future work creating a comprehensive framework to guide the building of prediction models.