Yahui Wang

Dr. Yahui Wang


Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

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

Representative Publications

View All Publications
22 October 2021 in Applied Sciences
Verified
Journal Article
Open access
PDF file
YWYahui Wang
Yahui Wang
WZWenxi Zhang
Wenxi Zhang
ZWZhou Wu
Zhou Wu
XKXinxin Kong
Xinxin Kong
HZHongxin Zhang
Hongxin Zhang
Laser speech measurement is a new sound capture technology based on Laser Doppler Vibrometry (LDV). It avoids the need for contact, is easily concealed and is ideal for remote speech acquisition, which has led to its wide-scale adoption for military and security applications. However, lasers are easily affected by complex detection environments. Thus, speckle noise often appears in the measured speech, seriously affecting its quality and intelligibility. This paper examines all of the characteristics of impulsive noise in laser measured speech and proposes a novel automatic impulsive noise detection and removal method. This method first foregrounds noise using decorrelation based on a linear prediction (LP) model that improves the noise-to-signal ratio (NSR) of the measured signal. This makes it possible to detect the position of noise through a combination of the average short-time energy and kurtosis. The method not only precisely locates small clicks (with a duration of just a few samples), but also finds the location of longer bursts and scratches (with a duration of up to a hundred samples). The located samples can then be replaced by more appropriate samples whose coding is based on the LP model. This strategy avoids unnecessary processing and obviates the need to compromise the quality of the relatively large fraction of samples that are unaffected by speckle noise. Experimental results show that the proposed automatic speckle noise detection and removal method outperforms other related methods across a wide range of degraded audio signals.

11 February 2022 in Applied Sciences
Verified
Journal Article
Open access
PDF file
YWYahui Wang
Yahui Wang
WZWenxi Zhang
Wenxi Zhang
ZWZhou Wu
Zhou Wu
XKXinxin Kong
Xinxin Kong
YWYongbiao Wang
Yongbiao Wang
HZHongxin Zhang
Hongxin Zhang
DNN-based Speech Enhancement (SE) models suffer from significant performance degradation in real recordings due to the mismatch between the synthetic datasets employed for training and real test sets. To solve this problem, we propose a new Generative Adversarial Network framework for Noise Modeling (NM-GAN) that creates realistic paired training sets by imitating real noise distribution. The proposed framework combines a novel 7-layer U-Net with two bidirectional long short-term memory (LSTM) layers that act as a generator to construct complex noise. NM-GAN generates enough recall (diversity) and precision (noise quality) in its samples through adversarial and alternate training, effectively simulating real noise, which is then utilized to compose realistic paired training sets. Extensive experiments employing various qualitative and quantitative evaluation metrics verify the effectiveness of the generated noise samples and training sets, demonstrating our framework’s capabilities.