A5100748467- Speech and Audio Processing
- Speech Recognition and Synthesis
- Emotion and Mood Recognition
- Advanced Computational Techniques and Applications
- Music and Audio Processing
- Education and Work Dynamics
- Advanced Neural Network Applications
- CCD and CMOS Imaging Sensors
- Face and Expression Recognition
- Machine Fault Diagnosis Techniques
- Advanced Data Processing Techniques
- Advanced Sensor and Control Systems
- Chinese history and philosophy
- Advanced Clustering Algorithms Research
- Smart Grid and Power Systems
- Advanced Decision-Making Techniques
- Face recognition and analysis
- Ideological and Political Education
- Machine Learning and ELM
- Infrared Target Detection Methodologies
- Sensor Technology and Measurement Systems
- Neural Networks and Applications
- Digital Communication and Language
- Network Security and Intrusion Detection
- Data Mining Algorithms and Applications
Ministry of Education of the People's Republic of China
2023
Nanjing Institute of Technology
2022
Southeast University
2008-2009
Huaqiao University
2008
Lanzhou University
2008
Harbin University
2004
DBSCAN is one of the most popular algorithms for cluster analysis. It can discover all clusters with arbitrary shape and separate noises. But this algorithm canpsilat choose parameter according to distributing dataset. simply uses global MinPts parameter, so that clustering result multi-density database inaccurate. In addition, when it used large databases, will cost too much time. For these problems, we propose GMDBSCAN which based on spatial index grid technique. An experimental evaluation...
When locating the rub-impact source of rotating machines by acoustic emission technology, precise time-difference arrival is always not available to calculate faults location due deteriorated signal correlation relationship among different sensors. This caused both dispersion phenomenon since includes various waveform modes while each type has velocity in complex rotor structure and intense noise influence inner device. To solve this problem, generalized cross correlated time delay...
In this paper, we propose a novel time-frequency joint learning method for speech emotion recognition, called Time-Frequency Transformer. Its advantage is that the Transformer can excavate global patterns in domain of signal while modeling local emotional correlations time and frequency respectively. For purpose, first design Time Frequency to capture between frames inside bands respectively, so as ensure integrity information both domains. Then, proposed mine through time-domain...
This paper proposes a modified Gaussian Mixed Model (GMM) with an embedded Time Delay Neural Network (TDNN). It integrates the merits of GMM which is generative model and TDNN discriminative model. digests timing information feature sequences, through transformation vectors it makes hypothesis variable independence that maximum likelihood needed more reasonable. are trained as whole by means probability. In process training, parameters updated alternately. Experiments show proposed improves...
This paper proposes a modified gaussian mixed model (GMM) with an embedded auto-associate neural network (AANN). It integrates the merits of GMM and AANN. AANN are trained as whole by means maximum likelihood. In process training, parameter updated alternately. reshapes distribution data improves similarity in one class. Experiments show that proposed system accuracy rate against baseline at all SNR, to 19%.
Dimension reduction is widely used in the domains of speech emotional recognition. Considering underlying time structure signals, we propose a local Fukunage-koontz transformation (LTFKT) to contain more discriminative information during The goal LTFKT maximize/ minimize eigenvalues covariance different classes simultaneously. It can be seen that FKT special case from analysis. results show method this paper improve recognition rate effectively.
Few‐shot segmentation is a challenging task due to the limited class cues provided by few of annotations. Discovering more from known and unknown classes essential few‐shot segmentation. Existing method generates mainly common intra new where similarity between support images query measured locate foreground regions. However, are not sufficient enough measure since one or mask cannot describe object with large variations. In this paper, we capture considering all in classes, i.e., only but...