Research Areas
Research Areas
Embedded Files
Deep Learning-Based Signal Processing
Deep Learning-Based Signal Processing
Recently, deep learning is widely used in the research of various applications. We are also working on applying deep learning technology to various signal processing problems to overcome the limitations of conventional techniques.
Our lab focuses on developing algorithms for speech, audio, and bio-signal processing based on deep neural networks (DNNs). Research topics include DNN-based speech enhancement, audio detection and classification, bio-signal detection for healthcare monitoring, and more.
Active Noise and Vibration Control
Active Noise and Vibration Control
Active Noise Cancellation (ANC) is a method of actively reducing undesired noise. ANC introduced a canceling ''antinoise'' wave using secondary sources (speakers). These secondary sources are interconnected through an electronic system using a specific signal processing algorithm for the particular cancellation scheme.
Our study aims to develop adaptive algorithms that can be used for narrowband and wideband ANC systems.
3-dimensional Audio Technologies
3-dimensional Audio Technologies
In the analysis and synthesis of 3D audio scenes, creating a 3D sound object without artifacts is crucial. Our team has been actively working in the MPEG audio group for the standardization of binaural 3D technologies.
Our research focuses on developing elementary algorithms for binaural systems, including compensation for the adverse effect of binaural downmixing.
Our research interest also includes filter design for the crosstalk cancellation and room equalization, bass enhancement for small-size speakers, and artificial reverberation for use in mobile devices, extracting the primary components from stereo audio for applications such as stereo upmixing, binaural auralization, and object-based 3D rendering.
Acoustic Echo Canceller
Acoustic Echo Canceller
Acoustic echo cancellation is essential for audio teleconferencing when full-duplex transmission of speech is necessary. The objective of the echo canceller is to detect and remove echoes as quickly and effectively as possible while minimizing any loss in voice quality. The performance of the echo canceller should be robust to the presence of background noise, and more importantly, to the presence of near-end speech. Thus, it is critical to handle the double-talk situation (a situation that both parties in the communication line are speaking at the same time).
Unlike the line echo, which occurs with relatively short lengths, the acoustic echo requires long-tap adaptive filters (several thousands-tap echo canceller is not uncommon), which raises computational and slow convergence problems.
Our research focuses on developing adaptive algorithms that can effectively solve the problems mentioned above.
Adaptive Feedback Cancellation for Hearing Aids
Adaptive Feedback Cancellation for Hearing Aids
Howling in hearing aids is caused by acoustic feedback from the receiver to the microphone. In practice, it not only limits the maximum usable gain but also degrades the sound quality of the hearing aids. A common approach to this well-known problem is to use an adaptive feedback canceller (AFC), in which the entire effect of feedback is eliminated by adaptively estimating the acoustic feedback path.
The most advanced feedback cancellation schemes monitor for feedback while the listener is wearing the hearing aid. The acoustic feedback is then estimated and eliminated by using an adaptive digital filter or notch filter. An important issue in feedback cancellation is that there exists a correlation between the input and output signals, which deteriorates the performance of the adaptive feedback canceller.
Our research focuses on developing fast but simple adaptive algorithms that can be implementable on battery-powered digital hearing aids.
Binaural Speech Enhancement
Binaural Speech Enhancement
By wearing hearing aids in both ears, it is much easier to understand speech in noisy environments. Further benefits for improving user comfort and speech intelligibility can be achieved by reducing the environmental noise. The real-life acoustic environment consists of various sound components such as desired speech signal, diffuse noise, and directional interference. Accordingly, the speech enhancement algorithms have to deal with all of the noise components in a unified framework.
Our research aims to develop unbiased target/noise PSD estimators for binaural digital hearing aids operating in a complex acoustic noisy environment.
Page updated
Google Sites
Report abuse