- Speech and Audio Processing
- Advanced Adaptive Filtering Techniques
- Speech Recognition and Synthesis
- Blind Source Separation Techniques
- Indoor and Outdoor Localization Technologies
- Direction-of-Arrival Estimation Techniques
- Music and Audio Processing
- Advanced X-ray Imaging Techniques
- Music Technology and Sound Studies
- Underwater Acoustics Research
- Construction Project Management and Performance
- Optical measurement and interference techniques
- Sparse and Compressive Sensing Techniques
- Image and Signal Denoising Methods
- Data Management and Algorithms
- Rough Sets and Fuzzy Logic
- Hearing Loss and Rehabilitation
- Neuroscience and Music Perception
- Advanced SAR Imaging Techniques
- Speech and dialogue systems
- Evaluation and Optimization Models
- Parallel Computing and Optimization Techniques
- Adaptive optics and wavefront sensing
- Structural Health Monitoring Techniques
- Analog and Mixed-Signal Circuit Design
Harbin Institute of Technology
2010-2025
University of Hong Kong
2018-2024
Chinese Academy of Sciences
2007-2022
Institute of Acoustics
2007-2022
Heilongjiang Provincial Hospital
2022
University of Chinese Academy of Sciences
2022
Hong Kong University of Science and Technology
2019
Energy Research Institute
2019
Technion – Israel Institute of Technology
2016
University of California, Davis
2016
A closed-loop nanoscale precision stage is integrated with an atomic force microscope to mechanically fabricate 3D nanostructures according predetermined designs, such as human face nanostructures, nanoline arrays of sine-wave and triangular nanodot sine-shaped, hemispheric, concave/convex nanopatterns in a controllable reproducible fashion.
Standing upon the intersection of traditional beamformers and deep neural networks, we propose a causal beamformer paradigm called Embedding Beamforming, two core modules are devised accordingly, namely EM BM. For EM, instead estimating spatial covariance matrix explicitly, 3-D embedding tensor is learned with network, where spatial-spectral discriminative information can be implicitly represented. BM, network directly leveraged to derive beamforming weights so as implement filter-and-sum...
Speech enhancement (SE) and neural vocoding are traditionally viewed as separate tasks. In this work, we observe them under a common thread: the rank behavior of these processes. This observation prompts two key questions: \textit{Can model designed for one task's degradation be adapted other?} \textit{Is it possible to address both tasks using unified model?} Our empirical findings demonstrate that existing speech models can successfully trained perform tasks, single model, when jointly...
The joint estimation of the spectrum, carrier, and direction arrival (DOA) is significant in radar, sonar, wireless communications, cognitive radio systems. traditional parameter measurement methods based on Shannon–Nyquist theorem require very high sampling rates, which put a lot pressure sampling, processing, storage devices. In this article, novel beamforming modulated wideband converter (BMWC) system for DOA proposed to improve robustness reduce structural complexity. From spatial signal...
Measurement of linear frequency modulation (LFM) signal is significant for radar, communication, and electronic reconnaissance fields. An LFM a wideband whose varies linearly with time, traditional measurement methods require very high sampling rates heavy processing to estimate parameters the signal. In this article, we propose multichannel cooperative (MCS) system based on finite rate innovation (FRI) theory sample real-valued pulse sequence (LFMPS). The MCS consists three parts:...
Recently, the quaternion-valued feedforward neural network (QFNN) has been developed to process three dimensional (3-D) and 4-D signals in quaternion domain, weight matrices bias vectors of QFNN were obtained based on backward propagation (QBP) method. However, it should be noted that QBP is a first-order gradient descent algorithm. The convergence speed usually slow may not very suitable nonstationary signals. To address this problem, widely linear unscented Kalman filter (WLQUKF) algorithm...
The code-shift keying (CSK) modulation method can achieve higher information transmission rates without changing the spread spectrum signal bandwidth. In order to optimise and demodulation of GNSS signals, in addition structure, binary phase-shift (BPSK) CSK signals using time-division multiplexing are proposed. A tracking based on BPSK-CSK is also proposed, which generates P-branch local codes by fast Fourier transform obtain code-slice spacings for E-branch L-branch codes. Then, tracked...
Whittaker-Shannon interpolation formula instead of narrowband filter (NF) is used to interpolate the signal more precisely in wideband direction-of-arrival (DOA) estimation problem when number samples highly limited. A novel algorithm named block FOCal Underdetermined System Solver (BFOCUSS) proposed solve this problem. The simulation results validate superior performance compared with other algorithms.
When applying the existing methods of establishing building information models, model users can make construction schedules and cost estimation easily. However, deviation modification schedule plans as well low data exchange efficiency would be seriously compromised under circumstances changing, real-time on-site information. This paper serves a foundation for large study over 4 years on based modelling context-aware technology. The aim this is to propose method develop prototype...
With the development of deep neural networks, automatic music composition has made great progress. Although emotional can evoke listeners' different auditory perceptions, only few research studies have focused on generating music. This paper presents EmotionBox -a music-element-driven generator based psychology that is capable composing given a specific emotion, while this model does not require dataset labeled with emotions as previous methods. In work, pitch histogram and note density are...
摘要: 先验信噪比是语音增强的关键参数。该文分析了几种典型的先验信噪比估计算法,并得到这几种算法的统一形式,最后提出了基于联合语音出现概率的先验信噪比估计算法。测试结果表明,该算法在不引入音乐噪声的同时,平均段信噪比提高和平均对数谱距离等客观评价指标,都优于其它算法。 关键词: 语音增强; 先验信噪比; 后验信噪比
Large graphs are increasingly prevalent in mobile networks, social traffic networks and biological networks. These often uncertain, where edges augmented with probabilities that indicates the chance to exist. Recently k-nearest neighbor search has been studied within field of uncertain graphs, but scalability efficiency issues not well solved. Moreover, solutions implemented on a single machine thus cannot fit large graphs. In this paper, we develop framework, called DURS, distribute into...
The method of super-resolution phase retrieval for Fourier phaseless measurement extends the signal model from discrete domain to a more realistic continuous domain. However, existing three-stage solving framework contains unnecessary redundancy, so its core steps are loosely connected and have high computational complexity. Aiming at this, new algorithms proposed following two key respectively, as achieve compact framework. First, stage auto-correlation function, non-redundant algorithm is...
Phaseless measurement is widely used in various fields, and phase retrieval a key step signal reconstruction of phaseless measurement. The occurrence outliers will cause the optimal solution traditional objective function to deviate from original signal, thereby reducing accuracy. This article modifies by introducing weight vector so that its still approximate when appear proposes specific implementation strategy for this idea under background Fourier retrieval. For vector, we design...