Xin Huang

Dr. Xin Huang


Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
17 March 2024 in Electronics
Verified
Journal Article
Open access
PDF file
CSChenjing Sun
Chenjing Sun
YZYi Zhou
Yi Zhou
XHXin Huang
Xin Huang
JYJichen Yang
Jichen Yang
XHXianhua Hou
Xianhua Hou
Speech emotion recognition poses challenges due to the varied expression of emotions through intonation and speech rate. In order to reduce the loss of emotional information during the recognition process and to enhance the extraction and classification of speech emotions and thus improve the ability of speech emotion recognition, we propose a novel approach in two folds. Firstly, a feed-forward network with skip connections (SCFFN) is introduced to fine-tune wav2vec 2.0 and extract emotion embeddings. Subsequently, ConLearnNet is employed for emotion classification. ConLearnNet comprises three steps: feature learning, contrastive learning, and classification. Feature learning transforms the input, while contrastive learning encourages similar representations for samples from the same category and discriminative representations for different categories. Experimental results on the IEMOCAP and the EMO-DB datasets demonstrate the superiority of our proposed method compared to state-of-the-art systems. We achieve a WA and UAR of 72.86% and 72.85% on IEMOCAP, and 97.20% and 96.41% on the EMO-DB, respectively.

11 April 2025 in Proceedings of the AAAI Conference on Artificial Intelligence
Verified
Journal Article
Open access
PDF file
HZHengde Zhu
Hengde Zhu
XKXiangyu Kong
Xiangyu Kong
WXWeicheng Xie
Weicheng Xie
XHXin Huang
Xin Huang
XHXilin He
Xilin He
SSSiyang Song
Siyang Song
In dyadic human-human interactions, individuals may express multiple different facial reactions in response to the same/similar behaviours expressed by their conversational partners depending on their personalised behaviour patterns. As a result, frequently-employed reconstruction loss-based strategies lead the training of previous automatic facial reaction generation (FRG) models to not only suffer from the 'one-to-many mapping' problem, but also fail to comprehensively consider the quality of the generated facial reactions. Besides, none of them considered such personalised behaviour patterns in generating facial reactions. In this paper, we propose the first adversarial FRG model training strategy which jointly learns appropriateness and realism discriminators to provide comprehensive task-specific supervision for training the target facial reaction generators, and reformulates the 'one-to-many (facial reactions) mapping' training problem as a 'one-to-one (distribution) mapping' training task, i.e., the FRG model is trained to output a distribution representing multiple appropriate/plausible facial reaction from each input human behaviour. In addition, our approach also serves as the first offline FRG approach that considers personalised behaviour patterns in generating of target individuals' facial reactions. Experiments show that our PerReactor not only largely outperformed all existing offline solutions for generating more appropriate, diverse and realistic facial reactions, but also is the first approach that can effectively generate personalised appropriate facial reactions.

8 October 2019 in IEEE Transactions on Vehicular Technology
Verified
Journal Article
Closed access
XHXin Huang
Xin Huang
FKFeng Ke
Feng Ke
RYRong Yu
Rong Yu
YLYuqin Liu
Yuqin Liu
YWYueqian Wu
Yueqian Wu
This paper focuses on 5G wireless dynamic network architecture and analyzes a novel paradigm for wireless network access, where certain classes of smart devices can become an access point (AP) temporarily anytime while connected to the Internet. Due to the unpredictable available/unavailable time and uninformed disconnections of the APs, users need a seamless, efficient and reliable schedule for switching between available APs. In this paper, we model the available/unavailable behavior of each AP as a two-state Markov chain, and then propose two switching policies: push-based preemptive switching and pull-based non-preemptive switching. For each policy, the process of user switching among the APs/BS in its accessing list is modeled as a Markovian process. Approximated statistical models of the two switching policies are provided, and stationary distributions of these processes can be derived. In addition, we propose a dual-charging mechanism where the APs charge the users by signaling transmission and data transmission. Based on the derived stationary distributions and proposed dual-charging mechanism, we propose an optimal priority assignment algorithm to constitute the access lists for preemptive and non-preemptive switching. Simulation results confirm the accuracy of the statis- tical models and verify the performance of the preemptive and non-preemptive switching policies. In the low mobility scenarios, preemptive switching would be a better option because it incurs higher capacity and utility for users, and higher profit for APs. In the high mobility scenarios, non-preemptive switching would be a better option because it incurs fewer switchings and higher profit for APs.

30 January 2024 in Electronics
Verified
Journal Article
Open access
PDF file
DWDe-Ming Wang
De-Ming Wang
婧胡婧 胡
婧 胡
XHXin Huang
Xin Huang
QZQinghua Zhong
Qinghua Zhong
Successive approximation register (SAR) analog-to-digital converters (ADC) have the advantages of a simple structure, low power consumption and a small area compared with other types of ADCs, and thus, high-performance SAR ADCs have always been a hot research topic in the industry. In this paper, a 12-bit SAR ADC design with calibration using a hybrid RC digital-to-analog converter(RC DAC) structure is proposed to improve the conversion accuracy of the ADC and reduce the circuit area at the same time. The analog supply voltage and reference voltage of the ADC are 3.3 V, and the digital supply voltage is 1.2 V. The ADC adopts a mixed digital–analog design scheme, in which the internal comparator, latch, DAC capacitor array, etc., are analog parts, and the rest of the SAR algorithms and calibration algorithms are all implemented in digital Verilog code, with a conversion accuracy of 0.8 mV and a calibration accuracy of 0.5 LSB. The ADC can be selectively calibrated, and the simulation shows that the accuracy of the calibrated ADC can be guaranteed to be within 2 LSB under a 14 MHz digital clock with a sampling rate of 1 MHz. After simulation at a sampling rate of 1 MHz and an input frequency of 244 Hz sine wave, the effective bit count of the ADC is 9.54 bits and the SFDR is 63.71 dB. The circuit consumes 1.78 mW with a 3.3 V supply voltage. The overall layout core area is 411 μm × 517 μm.