In this paper, we propose a new interactive controller for audio object localization based on spatially representative vector operations on a stereo mixed source. First, we develop the controller that enables the listener to intuitively operate of audio objects displayed on the touchscreen panel. Next, we assess the perceptual effects of localization and the sound quality of an audio object after performing individual operations with our controller via a subjective evaluation. The results of the experiments clarify that our controller enables the listener to change the localization of audio objects without sound degradation if the gain operation is not extreme.

抄録全体を表示

This paper proposes a rotation-robust detection method of images with resembling shapes using the local self-similarities. In particular, images do not necessarily share common visual properties such as colors, edges, and textures. Although the local self-similarity is effective for shape detection among such images, it lacks the robustness to image rotation, so that it is unable to match images of the same object in different orientations. We combine the center voting method with the self-similarity descriptor in order for giving the robustness to image rotation, where the orientation is assigned to each descriptor. After matching those oriented descriptors across images, the center voting is performed in the groups of the same angular difference between the assigned orientations of matched descriptors. The rotation robustness of the proposed method was proved by demonstrating experimental results.

抄録全体を表示

The restoration of bone-conducted speech is a very important issue that enables robust speech communication in extremely noisy environments. We proposed a method of blind restoration in our previous studies based on a scheme of linear prediction with a method of training and prediction based on the simple recurrent neural network. However, prediction based on neural networks is not suitable for training with large corpora, which is necessary for real applications. The over-training problem with simple recurrent neural networks makes it difficult to train various kinds of bone-conducted speech in one session. In addition, it is difficult to adapt the neural network model to bone-conducted speech in unknown noisy environments to build an open dataset restoration of bone-conducted speech. Thus, a method of training and prediction based on the Gaussian mixture model was used in this research, instead of a neural network. A method of re-estimating the residual ratio in the scheme of linear prediction is also proposed. We also investigated how the proposed method works to restore bone-conducted speech in extremely noisy environments. Objective and subjective evaluations were carried out to evaluate the improvements in sound quality and the intelligibility of restored speech. The results revealed that our proposed method outperformed previous methods in both human hearing and automatic speech recognition systems even in extremely noisy environments.

抄録全体を表示

Facial consciousncss and recognition have been studied for many years. However, these studies have only extracted the feature of facial impression. Some studies have treated the evaluation of facial impression. In those studies, liking, age, and gender were evaluated with respect to facial impression. Therefore, there are no details on whether we evaluate various general facial impressions in daily life. Moreover, the result of evaluation using those studied features depends on the facial-parts-extraction accuracy. Then, we use texture features for facial impression recognition in this study. The texture features can be extracted stably, because we need not specify their positions. We extract texture features in the mesh state. We can extract physical features by this technique without specifying the detailed location of features, unlike the conventional method. Finally, we show the effectiveness of our proposed method compared with previous studies.

抄録全体を表示

In wireless sensor networks, it is necessary to collect information from sensors using less energy. Therefore, it is proposed that representative sensors called a head collect information from neighbors and send it to a base station. The problem is how to decide heads autonomously. In a grid network, it is known that a method using pattern formation by a reaction-diffusion (RD) system is useful. However, energy efficiency is not so good, because locations of heads are fixed. In this paper we propose a method for deciding heads using pattern formation of a cellular automaton (CA). By computer simulation, we show that our method is less energy consumption than the method using the RD system.

抄録全体を表示

We present a method for automated sound quality evaluation of push-button sounds based on wavelet analysis. Push-button sounds are represented in the time-frequency plane using the continuous wavelet transform. For each sound, multiscale features identifying sound characteristics in terms of sound pressure and reverberation are compactly represented with triangular biorthogonal wavelets that are two-dimensional nonseparable wavelets defined on a triangular lattice. Together with an auditory impression of the sounds extracted by the semantic differential method, a preliminary experiment is then performed to evaluate automated recognition of push-button sounds using their features.

抄録全体を表示

In this study, we propose a method to analyse interindividual differences in electroencephalograms (EEGs) taken while a subject listened to music, which is based on the relationship of these EEGs to human personality. The frequencies of the EEGs that are analysed have components that contain both significant and immaterial information as well as different levels of importance. We express the different levels of importance through the weight values of frequencies using a real-coded genetic algorithm. Then, the EEG patterns, which are determined based on the evaluation of the impression on the music, are detected using the <i>k</i>-nearest neighbour method. We assume that differences between detection results for EEG patterns with the highest and lowest recognition accuracies show interindividual differences. Moreover, ego analysis based on psychological testing is used to analyse personality, and the ego score is determined using a questionnaire. Finally, we discuss the relationship between personality and interindividual differences observed in experimental EEGs. An interesting tendency of a person with a combined ego type is that he or she has a unique response to negative stimuli compared with that of positive stimuli.

抄録全体を表示

Various techniques based on the gradient descent method have been studied as training algorithms for neural networks. Neural network training poses data-driven optimization problems in which the objective function involves the summation of loss terms over a set of data to be modeled. For a given training data set, the gradient-based algorithm operates in one of two modes: online (stochastic) or batch. In this paper, a robust training algorithm is proposed, combining "online" mode with "batch" one. The validity of the proposed algorithm is demonstrated through computer simulations compared with the previous quasi-Newton based training methods.

抄録全体を表示