- Speech and Audio Processing
- Music and Audio Processing
- Advanced Adaptive Filtering Techniques
- Image and Signal Denoising Methods
- Speech Recognition and Synthesis
- Underwater Vehicles and Communication Systems
- Indoor and Outdoor Localization Technologies
- Hearing Loss and Rehabilitation
- Acoustic Wave Phenomena Research
- Underwater Acoustics Research
- Noise Effects and Management
- Blind Source Separation Techniques
- Advanced Data Compression Techniques
- Animal Disease Management and Epidemiology
- Influenza Virus Research Studies
- Internet of Things and Social Network Interactions
- Livestock and Poultry Management
- Vehicle Noise and Vibration Control
- Structural Health Monitoring Techniques
- Anomaly Detection Techniques and Applications
- Fluid Dynamics and Vibration Analysis
- Remote Sensing and Land Use
- Tactile and Sensory Interactions
- Energy and Environmental Systems
- Image Enhancement Techniques
Advanced Institute of Convergence Technology
2019-2023
Gwangju Institute of Science and Technology
2011-2021
National Forensic Institute
2017
Drones are widely utilized in various industries. Unfortunately, when a drone acquires sound through microphone, which is installed itself, flying and wind noises appear recorded signals. Therefore, it necessary to reduce such for enhancing the quality of signals UAV acoustic sensor networks. In this paper, we proposes noise reduction method using deep convolutional denoising autoencoder eliminating noises. The extract target source monaural audio separation. To do task, training dataset...
This paper proposes a sound event detection (SED) method in tunnels to prevent further uncontrollable accidents. Tunnel accidents are accompanied by crashes and tire skids, which usually produce abnormal sounds. Since the tunnel environment always has severe level of noise, accuracy can be greatly reduced existing methods. To deal with noise issue environment, proposed involves preprocessing acoustic signals classifier for detecting events tunnels. For preprocessing, non-negative tensor...
In this paper, a novel speech enhancement method based on hybrid machine-learning architecture consisting of U-Net and nonnegative matrix factorization (NMF) is proposed. The proposed attempts to take advantage both the accurate separation for known noise environments by adaptation unseen noises an NMF with online dictionary learning technique. To merge two different architectures, modified temporal activation layer (TAU-Net) jointly optimized models that represent universal noise. first...
An adaptive noise sensing method is proposed to improve the speech performance of speech-based applications operated over wireless sensor networks. The based on nonnegative matrix factorization (NMF), which consists and reduction. In other words, performed by adapting a priori basis NMF, estimated from signal, resulting in an adapted matrix. Subsequently, used for NMF decomposition noisy into clean background noise. signal then applied front-end applications. NMF-based reduction first...
This paper proposes a new non-stationary noise suppression method to reduce the mechanical generated when audio signals are recorded with digital camera. The proposed first utilizes non-negative matrix factorization (NMF) technique estimate spectrum of from noisy spectrum. After that, contaminated in signal is suppressed by multi-band spectral subtraction. In particular, NMF estimates frame-wise manner order for operate real-time. performance evaluated terms log-spectral distortion, cepstral...
This paper proposes a noise reduction method based on U-shaped neural network to effectively reduce wind noise. While the U-Net is developed for medical image segmentation, it constructed by using spectrograms of noisy input signals as feature, and trained estimate ideal ratio mask between pair clean target signals. The performance proposed measured in terms signal-to-distortion (SDR), signal-to-interference (SIR), signal-to-artifact (SAR). As result, shown that provides higher average SDR,...
In this paper, an audio denoising method is proposed for improving the quality of handheld recording devices. The reduces noise differently depending on block size in modified discrete cosine transform (MDCT) analysis coder. Specifically, a long performed by multi-band spectral subtraction (MBSS) with perceptually weighted scalefactor bands, while that short subband power scaling to maintain coherence previously-denoised block. order evaluate performance method, it first embedded into MPEG-2...
In this paper, a lightweight U-Net based monaural speech source separation method to implement high-quality functionality in an edge computing device having microphone is proposed. The proposed utilizes U-shaped neural networks segregate and interfering noises from input mixtures the time-frequency domain. To reduce sizes of suitable for real-time operation at resource-constrained edge-computing device, employs inception-like multi-lane dimensionality reduction module each convolutional...
A noncoherent low-frequency ultrasonic (LFU) communication system is proposed for near-field using commercial off-the-shelf (COTS) speakers and microphones. Since the LFU channel known to be a frequency-selective characteristic, basically designed by differential phase-shift keying (DPSK) modulation with forward error correction. In addition, automatic gain control of carrier frequency band over proposed. Then, in order optimize symbol length under realistic aerial acoustic channel,...
Reverberation degrades the speech quality and intelligibility, particularly for hearing impaired people. In an automatic recognition (ASR) system, a dereverberation technique, which removes reverberation, is widely employed as pre-processing to increase performance of ASR system. this paper, we compare CNN-based method by applying various vocoders. The U-Net architecture technique. WaveGlow, MelGAN, Griffin Lim are used Such vocoders play role in converting features into samples time domain,...
This paper proposes a new noise suppression method to reduce zoom generated when audio signals are recorded with digital camera. The proposed is based on multiband spectral subtraction that can suppress components of related reference zoom-noise in the modified discrete cosine transform domain. In particular, method, each frame classified as either or non-noise frame, and depending this classification, updated degree controlled. It shown from performance evaluation due zooming operation...
In augmented reality (AR), image markers are widely used for rendering virtual objects. However, the performance strongly depends on lighting environment and distance between marker a camera. Therefore, we propose an audio marker-based AR application using low frequency ultrasound (LFU) communication. The proposed method consisted of forward error correction (FEC) coding windowed differential phase shift keying (DPSK) modulation in order to make robust over noisy channel. successful...
In this paper, an artificial stereo extension method that creates stereophonic sound from a mono source is proposed. The proposed first trains deep neural networks (DNNs) model the nonlinear relationship between dominant and residual signals of channel. training stage, band-wise log spectral magnitude unwrapped phase both are utilized to nonlinearities each sub-band through architecture. From point, conducted by estimating signal corresponds input channel with trained DNN in domain....
This article proposes an efficient lossless compression method for underwater acoustic sensor array. The proposed first decides whether the input signal is coming from a normal or faulty sensor. In particular, such fault detection performed using root-mean-square crossing rate and zero rate. If determined to be faulty, then pre-processing technique applied prior make this meaningful further use. After that, multi-channel signals are encoded by MPEG-4 audio coding encoder, where indices of...
Over the last decade, highly pathogenic avian influenza (HPAI) has severely affected poultry production systems across globe. In particular, massive pre-emptive depopulation of all within certain distance raised concerns regarding animal welfare and food security. Thus, alternative approaches to reducing unnecessary depopulation, such as risk-based are demanded. This paper proposes a data-driven method generate rule table risk score for each farm identify preventive measures against HPAI. To...
Over the last decade, highly pathogenic avian influenza (HPAI) has severely affected poultry production systems across globe. In particular, massive pre-emptive depopulation of all within a certain distance raised concerns regarding animal welfare and food security. Thus, alternative approaches to reducing unnecessary depopulation, such as risk-based are demanded. This paper proposes data-driven method generate rule table risk score for each farm identify preventive measures against HPAI. To...
본 논문에서는 다음향(multisource) 환경에서의 음향 사건 검출 정확도를 높이기 위해 비음수 텐서 분해(nonnegative tensor factorization, NTF)와 은닉 마코프 모델(hidden Markov model, HMM)을 이용한 이중 채널 방법을 제안한다. 제안된 방법은 먼저 입력 신호들에 NTF 기법을 적용하여 얻은 각 별 이득을 활용하여 다수의 사건들을 검출한다. 그러고 나서, 이득에 의해 검출된 사건의 발생 여부를 검증하기 위하여 우도 가중치로 활용하는 HMM 기반의 우도비 검증을 수행한다. 방법의 평가하기 다양한 잡음과 사건간 중첩 밀도를 고려하는 다중 환경에 대한 F-measure를 측정하였고, 기존의 혼합 가우시안 모델 및 행렬분해 방법들과 비교하였다. 실험 결과, 방법이 기존 방법들에 비하여 모든 조건에서 높은 보였다.