- Speech and Audio Processing
- Speech Recognition and Synthesis
- Advanced Algorithms and Applications
- Blind Source Separation Techniques
- Machine Fault Diagnosis Techniques
- Voice and Speech Disorders
- Emotion and Mood Recognition
- Advanced Sensor and Control Systems
- Image and Signal Denoising Methods
- Music and Audio Processing
- Advanced Computational Techniques and Applications
- Digital Media and Visual Art
- Simulation and Modeling Applications
- Industrial Vision Systems and Defect Detection
- Face recognition and analysis
- Advanced Measurement and Detection Methods
- Face and Expression Recognition
- Fault Detection and Control Systems
- Education and Critical Thinking Development
- Structural Health Monitoring Techniques
- Financial Reporting and Valuation Research
- Advanced Image Fusion Techniques
- Neuroscience, Education and Cognitive Function
- EEG and Brain-Computer Interfaces
- Corporate Finance and Governance
Guangxi Normal University
2012-2024
Peking University
2023-2024
Guangdong University Of Finances and Economics
2023
Guilin University of Electronic Technology
2001-2022
Sichuan University
2022
West China Second University Hospital of Sichuan University
2022
Nankai University
2022
Guangxi University of Science and Technology
2015-2019
Shaanxi Normal University
2013
Kunming University of Science and Technology
2006-2010
Newer antiepileptic drugs such as levetiracetam, lacosamide, topiramate, gabapentin, oxcarbazepine, lamotrigine, and zonisamide are prescribed by physicians for the treatment of epilepsy. These also associated with a series eye disorders. However, very few studies have systemically compared disorders newer AEDs in large sample patients diagnosed epilepsy.This study aimed to evaluate association between several AEDs, examine differences frequency adverse events across individual through data...
The present study investigated the electrophysiological correlates of morphological processing in Chinese compound word reading using a delayed repetition priming paradigm. Participants were asked to passively view lists two-character words containing prime-target pairs separated by few items. In Whole Word condition, prime and target same real (e.g., , manager-manager). Constituent swapped terms their constituent position former is pseudo-word later means nurse). Two ERP components...
Speech1 Emotion Recognition SER uses the Berlin EMO-DB database, seven emotions. Traditional emotional features and their statistics are used in SER. Two improved Mel Frequency Cepstrum Coefficients MFCC added to this experiment, which extract parameters from energy curve fundamental frequency curve, that EEMFCC F0MFCC, using Support Vector Machines SVM as recognition machine, we obtained highest average rate of 85.37% for categories 100% sad.
Basic structure of latch-type SRAM sense amplifier is analyzed and advantages disadvantages are compared in this paper, then an improved presented. On basis, a new proposed, which can access data fast for low voltage power application. The simulation results show that has advantage over the conventional one high-speed low-power, based on industry standard 1.0V/65 nm CMOS technology.
Marginal spectrums of cough sound and non-cough are obtained through Hilbert-Huang Transform. Speech signals five different sounds, namely throat clearing, sigh voice, shouting speech voice laughter, analyzed contrastively focusing on the characteristics marginal spectrum. Then SECC is extracted for recognition. The paper mainly recognizes characteristic parameters MFCC coefficient by using Continuous Hidden Markov Model(CHMM). recognition result shows that parameter based Hilbert spectrum...
In order to solve the problem of mode mixing and aliasing arising from speech decomposition, this paper proposes a signal decomposition method based on Variational Mode Decomposition (VMD): Decomposition-Frequency Shifting, VMD-FS). The takes advantage VMD's good extraction fundamental frequency signal, sets specific carrier parameters shift lower frequency, then applies iterative methods VMD decompose in obtain true IMFs that make up signal. Through experiments real signals, it is...
To alleviate the problem of severe degradation speaker recognition performance because phoneme variability between training and testing speech data, in text-independent system. The paper proposed a (TI) identification method that suppresses phonetic information by subspace method, Probabilistic Principle Component Analysis (PPCA) is utilized to construct these subspaces. Firstly, covariance matrix was obtained from large feature then projection using EM algorithm. In it assumed with variance...
In order to solve the problem of face recognition in natural illumination, a new algorithm using Eigenface-Fisher Linear Discriminant (EFLD) and Dynamic Fuzzy Neural Network (DFNN) is proposed this paper, which can dimension feature, deal with classification easily. we use EFLD model extract will be considered as input DFNN. And DFNN implemented classifier classification. The has been tested on ORL database. experiment results show that reduces feature finds best subspace for human face. by...
Speech emotion recognition (SER) is always challenging because of factors such as emotional corpus, acoustic features and SER modeling. based on deep learning are limited to using a spectrogram or handcrafted input, but cannot capture enough the defects information, this paper proposes feature fusion method Bidirectional Long Short-Term Memory (BLSTM) Convolutional Neural Networks (CNN) study richer features, combining context spatial features. Statistical used input BLSTM network, speech...
Depression has become a common mental disorder that plagues more and people. This paper uses speech signals to study method for predicting the degree of depression help clinicians judge in patients. In this paper, autoencoder model based on Bidirectional Gated Recurrent Unit (BiGRU) is proposed extract deep features, with original as network input, signal after cepstrum separation training target. model, we take input network, homomorphic target model. The long-term features extracted by...
Speech emotion is divided into four categories, Fear, Happy, Neutral and Surprise in this paper. Traditional features their statistics are generally applied to recognize speech emotion. In order quantify each feature’s contribution recogni-tion, a method based on the Back Propagation (BP) neural network adopted. Then we can obtain optimal subset of features. What’s more, two new characteristics emotion, MFCC feature extracted from fundamental frequency curve (MFCCF0) amplitude perturbation...
A modified pitch period detection (MPPD) algorithm is proposed in this paper. It based on the redefinition of instantaneous energy density level Hilbert-Huang transform, called as frequency weighted FIE(t). Two examples are employed to illuminate physical meaning new definition. MPPD used experiments with 78 normal and 70 abnormal vowel /a:/ voices, experimental results show that provides more robust performance makes better voice detection.
This paper proposes a new method for pitch extraction, especially useful pathological voice, by using the wavelet transform in frequency domain, and disregarding upper half signal. In this way benefits of discrete dyadic are combined performance cepstral is dramatically improved. can also be used robust extraction noisy environments.
In this paper we investigate the frequency resolution of empirical mode decomposition (EMD) and explore means to improve it through a series numerical experiments. Using compound signal composed two pure tones, find marginal ratio at which EMD can no longer separate them into different intrinsic functions (IMFs). Based on observation determine optimal for masking signal, application not only alleviate mixing but also enhance EMD.
A lightweight and fast skew detection recognition method is proposed to address low accuracy, slow speed, the inability detect skewed spray codes on complex background packaging. Built upon YOLOv5-obb network, approach utilizes Ghost module backbone reducing parameters computations. The introduction of Slim-neck feature fusion network structure in neck further simplifies model while enhancing accuracy. SimAM added both improve overall rates. In post-processing, a for merging scene text...
In response to the issues of low recognition accuracy in silent electromyographic facial action speech reconstruction tasks, this paper combines residual network (ResNet) and Transformer model design a Res based on ResNet structure network. The consists three connected structures several modules, features are extracted using ResNet, converts signal into Mel frequency spectrum 80 bands, sends HiFiGAN for reconstruction, ultimately obtaining audible under action. addition, our work also...