Seminar on Audio Perception and Evaluation
Audio perception plays an important role in system design and evaluation. The seminar focuses on various aspects from quantification of sound quality, psychoacoustic modeling and audiovisual effects.
RegistrationTo register for this event you must be logged in as a member of ASIP-NET.
Time and place
February 15, 2007, 13.00-17.10
GN ReSound, Lautrupbjerg 7, 2750 Ballerup, Denmark
Welcome by Lars Spicker Olesen, GN ReSound
Objective quantification of sound quality using descriptive analysis
Head of Research, Dr. Søren Bech, Bang&Olufsen A/S, Denmark
Descriptive Analysis is a technique that allows the researcher to obtain a complete sensory description of a given product and further to establish the importance of the individual attributes in relation to overall preference or quality. Descriptive Analysis includes a number of different techniques and was originally developed in the food industry. B&O and Philips Research in a joint research project further developed the technique for use in quantification of image quality and B&O, Nokia, BBC, and University of Surrey in another project developed the technique to be used for quantification of sound quality. The technique is now used as a routine tool for quantification of sound and image quality. The talk will introduce the technique and give an example of its use. The technique and its use (including a number of examples) in audio evaluation can be found in the recently published book by Bech & Zacharov "Perceptual Audio Evaluation - Theory, Method and Application".
Psychoacoustic models and their application in modern media systems
The development of psychoacoustics as a scientific enterprise in the past 100 years has been strongly linked with technological developments. For example, the influential research program in speech and hearing at the research laboratories of the "Bell Telephone System", started around 1914, was motivated by the technical and economical need to improve the design of telephone systems. With the advance of digitization of all areas of storage, modification and reproduction of audio siganls, there are also increasing possibilities for applying advanced perceptual models in the design and evaluation of modern media systems. The most widely experienced application area is digital audio coding, based on so-called "perceptual coding". Early realizations of such coding algorithms, like the MPEG1 standard which includes the well-known MP3 format, mainly exploited spectral masking properties of the auditory system: The error signal caused by amplitude quantization was spectrally shaped in such a way as to be optimally masked by the audio signal itself. Such algorithms can be improved by taking spectral and temporal integration properties of the hearing system into account. More recent approaches to obtain a high perceptual quality at even lower bit rates try to model the audio signal by a parametrical description using signal prototypes, like transients, sinusoidal components and noise. Also spatial properties of stereo and multichannel signals have succesfully been parameterized. Such signal representations do not only allow a strong bit rate reduction, but they also offer new opportunities for signal manipulation, like pitch shifting, duration change and spatial repositioning of sound sources. A second area of interest is automatic analysis of audio signals, as it is needed for, e.g., automatic classification and music transcription. This type of analysis is not only studied for applications in modern media systems, but it is also applied in hearing aids to automatically classify the acoustic environment which is then used to control the hearing aid settings.
University Flash - ongoing university projects
Coffee break, networking and posters
Assoc. Prof., Dr. Søren Kyllingsbæk, Copenhagen University, Denmark
Mathematical modeling has become an increasingly important tool in understanding cognitive processes. At the Center of Visual Cognition (CVC), Department of Psychology at the University of Copenhagen sophisticated computational models of visual cognition have been developed over the last 30 years. More specifically we have developed a Theory of Visual Attention (TVA) which is specified both at the behavioral as well as at the single cell level in the brain, thus bridging cognition and neurophysiology (Bundesen, 1990; Bundesen, Habekost, & Kyllingsbæk, 2005). I will briefly describe the theoretical framework of TVA and different applications (e.g. neurological deficits following brain damage). Recently, a collaborative Center for Computational Cognitive Modeling has been formed between the CVC and the Intelligent Signal Processing group at the IMM, Technical University of Denmark where we aim to extend our efforts into computational modeling of auditory perception and attention.
Postdoc, Dr. Tobias Andersen, Technical University of Denmark
Auditory perception is not isolated from the other senses. Intuitively, we experience the auditory and visual senses as separate but what we see can, in fact, categorically change what we hear and vice versa. Such effects of audiovisual integration occur in speech perception, numerocity judgments and signal detection. They depend strongly on stimulus configurations and may also depend on the cognitive state of the observer. From studies of these dependencies, we can begin to describe the nature of audiovisual integration in human perception.
The seminar is supported by the Signal Processing Chapter of the IEEE Denmark Section.