Background One of the most critical issues in human-computer interaction applications is recognizing human emotions based on speech.In recent years,the challenging problem of cross-corpus speech emotion recognition(SE...Background One of the most critical issues in human-computer interaction applications is recognizing human emotions based on speech.In recent years,the challenging problem of cross-corpus speech emotion recognition(SER)has generated extensive research.Nevertheless,the domain discrepancy between training data and testing data remains a major challenge to achieving improved system performance.Methods This paper introduces a novel multi-scale discrepancy adversarial(MSDA)network for conducting multiple timescales domain adaptation for cross-corpus SER,i.e.,integrating domain discriminators of hierarchical levels into the emotion recognition framework to mitigate the gap between the source and target domains.Specifically,we extract two kinds of speech features,i.e.,handcraft features and deep features,from three timescales of global,local,and hybrid levels.In each timescale,the domain discriminator and the feature extrator compete against each other to learn features that minimize the discrepancy between the two domains by fooling the discriminator.Results Extensive experiments on cross-corpus and cross-language SER were conducted on a combination dataset that combines one Chinese dataset and two English datasets commonly used in SER.The MSDA is affected by the strong discriminate power provided by the adversarial process,where three discriminators are working in tandem with an emotion classifier.Accordingly,the MSDA achieves the best performance over all other baseline methods.Conclusions The proposed architecture was tested on a combination of one Chinese and two English datasets.The experimental results demonstrate the superiority of our powerful discriminative model for solving cross-corpus SER.展开更多
The three-dimensional structure of a biomolecule rather than its one-dimensionM sequence determines its biological function. At present, the most accurate structures are derived from experimental data measured mainly ...The three-dimensional structure of a biomolecule rather than its one-dimensionM sequence determines its biological function. At present, the most accurate structures are derived from experimental data measured mainly by two techniques： X-ray crystallog- raphy and nuclear magnetic resonance （NMR） spec- troscopy. Because neither X-ray crystallography nor NMR spectroscopy could directly measure the positions of atoms in a biomolecule, algorithms must be designed to compute atom coordinates from the data. One salient feature of most NMR structure computation algorithms is their reliance on stochastic search to find the lowest energy conformations that satisfy the experimentally- derived geometric restraints. However, neither the cor- rectness of the stochastic search has been established nor the errors in the output structures could be quantified. Though there exist exact algorithms to compute struc- tures from angular restraints, similar algorithms that use distance restraints remain to be developed. An important application of structures is rational drug design where protein-ligand docking plays a crit- ical role. In fact, various docking programs that place a compound into the binding site of a target protein have been used routinely by medicinal chemists for both lead identification and optimization. Unfortunately, de- spite ongoing methodological advances and some success stories, the performance of current docking algorithms is still data-dependent. These algorithms formulate the docking problem as a match of two sets of feature points. Both the selection of feature points and the search for the best poses with the minimum scores are accomplished through some stochastic search methods. Both the un- certainty in the scoring function and the limited sam- pling space attained by the stochastic search contribute to their failures. Recently, we have developed two novel docking algorithms： a data-driven docking algorithm and a general docking algorithm that does not rely on experimental data. Our algorithms search the pose space exhaustively with the pose space itself being limited to a set of hierarchical manifolds that represent, respectively, surfaces, curves and points with unique geometric and energetic properties. These algorithms promise to be es- pecially valuable for the docking of fragments and small compounds as well as for virtual screening.展开更多
基金the National Nature Science Foundation of China(U2003207,61902064)the Jiangsu Frontier Technology Basic Research Project(BK20192004).
文摘Background One of the most critical issues in human-computer interaction applications is recognizing human emotions based on speech.In recent years,the challenging problem of cross-corpus speech emotion recognition(SER)has generated extensive research.Nevertheless,the domain discrepancy between training data and testing data remains a major challenge to achieving improved system performance.Methods This paper introduces a novel multi-scale discrepancy adversarial(MSDA)network for conducting multiple timescales domain adaptation for cross-corpus SER,i.e.,integrating domain discriminators of hierarchical levels into the emotion recognition framework to mitigate the gap between the source and target domains.Specifically,we extract two kinds of speech features,i.e.,handcraft features and deep features,from three timescales of global,local,and hybrid levels.In each timescale,the domain discriminator and the feature extrator compete against each other to learn features that minimize the discrepancy between the two domains by fooling the discriminator.Results Extensive experiments on cross-corpus and cross-language SER were conducted on a combination dataset that combines one Chinese dataset and two English datasets commonly used in SER.The MSDA is affected by the strong discriminate power provided by the adversarial process,where three discriminators are working in tandem with an emotion classifier.Accordingly,the MSDA achieves the best performance over all other baseline methods.Conclusions The proposed architecture was tested on a combination of one Chinese and two English datasets.The experimental results demonstrate the superiority of our powerful discriminative model for solving cross-corpus SER.
文摘The three-dimensional structure of a biomolecule rather than its one-dimensionM sequence determines its biological function. At present, the most accurate structures are derived from experimental data measured mainly by two techniques： X-ray crystallog- raphy and nuclear magnetic resonance （NMR） spec- troscopy. Because neither X-ray crystallography nor NMR spectroscopy could directly measure the positions of atoms in a biomolecule, algorithms must be designed to compute atom coordinates from the data. One salient feature of most NMR structure computation algorithms is their reliance on stochastic search to find the lowest energy conformations that satisfy the experimentally- derived geometric restraints. However, neither the cor- rectness of the stochastic search has been established nor the errors in the output structures could be quantified. Though there exist exact algorithms to compute struc- tures from angular restraints, similar algorithms that use distance restraints remain to be developed. An important application of structures is rational drug design where protein-ligand docking plays a crit- ical role. In fact, various docking programs that place a compound into the binding site of a target protein have been used routinely by medicinal chemists for both lead identification and optimization. Unfortunately, de- spite ongoing methodological advances and some success stories, the performance of current docking algorithms is still data-dependent. These algorithms formulate the docking problem as a match of two sets of feature points. Both the selection of feature points and the search for the best poses with the minimum scores are accomplished through some stochastic search methods. Both the un- certainty in the scoring function and the limited sam- pling space attained by the stochastic search contribute to their failures. Recently, we have developed two novel docking algorithms： a data-driven docking algorithm and a general docking algorithm that does not rely on experimental data. Our algorithms search the pose space exhaustively with the pose space itself being limited to a set of hierarchical manifolds that represent, respectively, surfaces, curves and points with unique geometric and energetic properties. These algorithms promise to be es- pecially valuable for the docking of fragments and small compounds as well as for virtual screening.
