Harmonic/percussive source separation (HPSS) is an important tool for analyzing and processing audio signals. The standard approach to HPSS takes advantage of the structural difference of sinusoidal and percussive components, calledanisotropic smoothness, in magnitude spectrograms. However, the existing methods disregard phase of the spectrograms and/or approximate the problem, which naturally limits the upper bound of the performance of HPSS. In this letter, we propose a novel approach to HPSS that regards phase without the approximation. The proposed method introduces an auxiliary variable that acts as an adaptive weight of a weighted energy minimization problem, which enables us to apply smoothing on magnitude of complex-valued spectrograms. Compared to the existing methods, the proposed method can obtain separated components having better magnitude and phase by simultaneously handling them.

In this paper, we propose a restoration method of time-varying graph signals, i.e., signals on a graph whose signal values change over time, using deep algorithm unrolling. Deep algorithm unrolling is a method that learns parameters in an iterative optimization algorithm with deep learning techniques. It is expected to improve convergence speed and accuracy while the iterative steps are still interpretable. In the proposed method, the minimization problem is formulated so that the time-varying graph signal is smooth both in time and spatial domains. The internal parameters, i.e., time domain FIR filters and regularization parameters, are learned from training data. Experimental results using synthetic data and real sea surface temperature data show that the proposed method improves signal reconstruction accuracy compared to several existing time-varying graph signal re- construction methods.

Squat training is highly effective for improving lower limb muscle function. However, when this type of training is performed alone without a personal trainer or physical therapist, it can lead to inefficiencies and injuries among trainees. Thus, we aim to develop a system that provides visual and real-time feedback on the correct posture of a lone squat trainee using pose estimation. In this study, we propose a function that demonstrates the correct posture for each squatting discipline in the form of a line-segment posture representation and warns the trainee when he/she assumes an incorrect posture. Squatting motion in the sagittal plane was captured using a camera connected to a personal computer, and the coordinates of the acromion, hip, knee, and ankle joints were detected using MoveNet. The joint angles were calculated from the detected coordinates and the correct posture was determined according to the individual's body shape. The color of the formed line-segment posture representation changed to indicate the difference between the trainees’ actual and correct postures. In future research, we plan to assess the displacement of key points detected by MoveNet using optical technologies to confirm the reliability of this system’s pose estimation. Afterwards, we will evaluate the effectiveness of this system in determining whether a trainee can move appropriately.

In this paper, we propose a system that supports operators who provide services to customers using teleoperated robots. We observed that unprofessional or lazy operators of teleoperated robots are a risk for businesses as they are likely to speak in ways that are inappropriate for customer services. The proposed system lets competent operators talk freely to customers and thus provide high quality service. For subpar operators, the proposed system filters inappropriate utterances to improve the service they provide. We conducted a user study with 21 participants to compare the proposed support system to a baseline system where operators talk freely to customers. For subpar operators, the quality of the service is significantly higher in terms of perceived politeness and reported customer satisfaction when using the proposed support system compared to when using the baseline system. For competent operators, we found no significant differences in the quality of the service between the two systems.

Sparse time-frequency (T-F) representations have been an important research topic for more than several decades. Among them, optimization-based methods (in particular, exten-sions of basis pursuit) allow us to design the representations through objective functions. Since acoustic signal processing uti-lizes models of spectrogram, the flexibility of optimization-based T-F representations is helpful for adjusting the representation for each application. However, acoustic applications often require models of magnitude of T-F representations obtained by discrete Gabor transform (DGT). Adjusting a T-F representation to such a magnitude model (e.g., smoothness of magnitude of DGT coefficients) results in a non-convex optimization problem that is difficult to solve. In this paper, instead of tackling difficult non-convex problems, we propose a convex optimization-based framework that realizes a T-F representation whosemagnitudehas characteristics specified by the user. We analyzed the prop-erties of the proposed method and provide numerical examples of sparse T-F representations having, e.g., low-rank or smooth magnitude, which have not been realized before.

In this paper, we propose a parameter selection method for graph learning. Graph learning, a technique of learning graphs from observations, is required in many applications, e.g., classification, prediction, and clustering. However, there is no established method to determine hyperparameters that control the strength of the regularization reflecting prior knowledge. To resolve the problem, we consider a model selection criterion for the graph learning problem based on Laplacian constrained Gaussian Markov random field. The proposed criterion is the value based on model evidence, which is used for model selection in Bayesian statistics. It can be estimated by averaging the negative log-likelihood over the posterior distribution of a graph learning model. To compute this criterion, we present an efficient sampler of the posterior distribution. In the experiment with random graphs, we demonstrate that the proposed method can select hyperparameters having a good trade-off between F-measure and relative error.

Finding important edges in a graph is a crucial problem for various research fields such as network epidemics, signal processing, machine learning, and sensor networks. In this paper, we tackle the problem based on sampling theory on graphs. We convert the original graph to a line graph where its nodes and edges, respectively, represent the original edges and the connections between the edges. We then perform node sampling of the line graph based on the edge smoothness assumption: This process selects the most important edges in the original graph. We present a general framework of edge sampling based on graph sampling theory and we also reveal a theoretical relationship between the original and line graphs. Experimental results in synthetic graphs validate the effectiveness of our approach against some alternative edge selection methods.

We propose a design method of sampling matrices for graph signals that guarantees perfect recovery for arbitrary graph signal subspaces. When the signal subspace is known, perfect reconstruction is always possible from the samples with an appropriately designed sampling matrix. However, most graph signal sampling methods so far design sampling matrices based on the bandlimited assumption and sometimes violates the perfect reconstruction condition for the other signal models. In this paper, we formulate an optimization problem for the design of the sampling matrix that guarantees perfect recovery, thanks to a generalized sampling framework for standard signals. In experiments with various signal models, our sampling matrix presents better reconstruction accuracy both for noiseless and noisy situations.

In this paper, we propose a deep algorithm unrolling (DAU) based on a variant of the alternating direction method of multiplier (ADMM) called Plug-and-Play ADMM (PnP-ADMM) for denoising of signals on graphs. DAU is a trainable deep architecture realized by unrolling iterations of an existing optimization algorithm which contains trainable parameters at each layer. We also propose a nested-structured DAU: Its submodules in the unrolled iterations are also designed by DAU. Several experiments for graph signal denoising are performed on synthetic signals on a community graph and U.S. temperature data to validate the proposed approach. Our proposed method outperforms alternative optimization- and deep learning-based approaches.

We propose a method for graph learning from spatiotemporal measurements. We aim at inferring time-varying graphs under the assumption that changes in graph topology and weights are sparse in time. The problem is formulated as a convex optimization problem to impose a constraint on the temporal relation of the time-varying graph. Experimental results with synthetic data show the effectiveness of our proposed method.

This paper proposes a method to generate synthesized underwater images from clean RGB-D images taken on the ground. It is beneficial for training a deep neural network for underwater image restoration (UWIR), and also for measuring the performances among UWIR methods. The underwater images are synthesized on the modeling of an accurate degradation process with the consideration of absorption and scattering as well as ten water types. The water types result in different attenuation coefficients, i.e., different synthesized images. In the experimental results, it is validated that our method successfully synthesizes underwater images, and presents a state-of-the-art performance for UWIR by utilizing our synthesized images for the training of deep learning-based UWIR.