Publikationen

The field of Industrial Sound Analysis (ISA) aims to automatically identify faults in production machinery or manufactured goods by analyzing audio signals. Publications in this field have shown that the surface condition of metal balls and different types of bulk materials (screws, nuts, etc.) sliding down a tube can be classified with a high accuracy using audio signals and deep neural networks. However, these systems suffer from domain shift, or dataset bias, due to minor changes in the recording setup which may easily happen in real-world production lines. This paper aims at finding methods to increase robustness of existing detection systems to domain shift, ideally without the need to record new data or retrain the models. Through five experiments, we implement a convolutional neural network (CNN) for two publicly available ISA datasets and evaluate transfer learning, data normalization and data augmentation as approaches to deal with domain shift. Our results show that while supervised methods with additional labeled data are the best approach, an unsupervised method that implements data augmentation with adaptive normalization is able to improve the performance by a large margin without the need of retraining neural networks.

The aim of auditory augmented reality is to create an auditory illusion combining virtual audio objects and scenarios with the perceived real acoustic surrounding. A suitable system like position-dynamic binaural synthesis is needed to minimize perceptual conflicts with the perceived real world. The needed binaural room impulse responses (BRIRs) have to fit the acoustics of the listening room. One approach to minimize the large number of BRIRs for all source-receiver relations is the synthesis of BRIRs using only one measurement in the listening room. The focus of the paper is the evaluation of the spatial audio quality. In most conditions differences in direct-to-reverberant-energy ratio between a reference and the synthesis is below the just noticeable difference. Furthermore, small differences are found for perceived overall difference, distance, and direction perception. Perceived externalization is comparable to the usage of measured BRIRs. Challenges are detected to synthesize more further away sources from a source position that is more close to the listening positions.

In recent years, high dynamic range (HDR) has been improved enormously. The capability of cameras and displays to reproduce small differences in luminance levels is constantly growing. However, we are still dealing with a limitation of the human visual system (HVS) known as the simultaneous contrast range (SCR). Compared to earlier studies, this paper focuses on real-world scenarios for evaluating the SCR. In natural images, bright highlights, especially in HDR, can limit the eyes' sensitivity to small differences in surrounding dark areas. This paper describes a test-image set developed as part of current research activities by the authors to measure the relation between the perceived SCR and the following four significant parameters: the distance, or rather, the viewing angle; the size of the bright highlight; the luminance of the highlight; and the ambient light. As a result, a mathematical formula is given that can help to evaluate and improve HDR viewing experiences as well as standard dynamic range downconversions.