Ming Yang: Soundscape, emotion, and machine learning
Okt 18 um 11:00 – 12:00

Everything that makes up our physical surroundings affects our feelings and emotional states. It further affects our behaviours, cognitive performances, and mental/physical health. In the study fields of soundscape, psychology, as well as neurobiology, it has been widely recognised that one’s emotional states are influenced by the external physical environment through multiply sensory perceptions and information cognition, and depend on many other factors such as one’s activity, mind state, and personality. The presentation will report some important parts of my previous PhD and postdoctoral research in the auditory sensation, cognition and emotion of soundscape/environment. It includes the differences in auditory sensations between generally pleasant and unpleasant environmental sounds in terms of natural and anthropogenic sounds, computer recognition of environmental sounds through signal processing (auditory sensations) and machine learning, and prediction of emotional assessment of environmental sounds based on semantic (cognitive) and psychoacoustic factors.

Chendi Zhu: Untersuchung der Messunsicherheit von Außenohrübertragungsfunktionen mittels Pol-Nullstellen-Analyse
Dez 6 um 11:30 – 12:00

Außenohrübertragungsfunktionen (engl. Head-related transfer functions, HRTFs) beschreiben den Filterprozess des am Ohr eintreffenden Schalls durch Reflexionen, Brechungen, Beugungen, Interferenzen und Resonanzen. Die HRTFs besitzen einen richtungsabhängigen Anteil und unterscheiden sich außerdem von Person zu Person.

Ein neues schnelles Messsystem für HRTFs wird am Institut für Technische Akustik verwendet. Viele Ursachen können jedoch die Messungen individueller HRTFs beeinflussen, z.B. Probandenbewegung.

Ziel dieser Arbeit ist es, die Messgenauigkeit bzw. die Wiederholbarkeit zu untersuchen. Der Schwerpunkt ist die Analyse der Veränderung des Frequenzspektrums für jeweils die gleiche Richtung bei wiederholten HRTF-Messungen für jeweils eine Person. Dazu wird eine Pol-Nullstellen-Analyse verwendet, um die Maxima und Minima vom Frequenzspektrum zu approximieren. Verschiedene Verfahren können dafür eingesetzt werden, beispielsweise mit dem Ziel, die logarithmische Differenz zwischen der rekonstruierten und der gemessenen HRTF zu minimieren. Um Fehler bzw. Unterschiede zwischen HRTFs zu bewerten, wird in dieser Arbeit ein Datensatz von gemessenen HRTFs verwendet.

David Kliesch: Uncertaintiy discussion of spatial sound field changes in auditoria
Dez 20 um 11:00 – 12:00

Rooms are often characterised by room acoustic metrics which help to categorize them by required standards. Through a previous work sound elds have been sampled in set grids and datasets have been generated. It has been shown that as an example the change in the clarity index C80 between measurement positions next to each other may already exceed the just noticeable threshold of 1dB. As a common experience it is unlikely to perceive a di erence while for example moving along a row of seats in a concert hall. The reproducibility of measurements is problematic since small di erences in measurement positions may already lead to an uncertainty. GUM, the guide to expression of uncertainties in measurement, o ers a framework which is used to estimate a model function accounting for a number of in uence quantities.
According to the formulation stage of the GUM framework one approach to estimate a model function is to nd a mean value which associates any given input distance between two microphones with an expected variation in a discussed metric. Using the dataset any sampling position can be compared with each other to evaluate metrics as a function of frequency and bandwidth. The sampling locations are uncertain themselves and have to be accounted for in a re ned model function. The distance between the microphones serve as input quantity while the average change of the examined metric is estimated as output quantity. In reference to GUM’s calculation stage the model function can be used to describe the propagation of measurement uncertainties. Since the measurement function may be not strictly linear the resulting output distribution is distorted. Using Monte-Carlo simulations these newly distorted  distribution functions can be estimated. The conclusion is that boundaries for the measurement accuracy in dependence of a tolerable uncertainty can be established.