Sound and Vision

Don’t you wonder sometimes

’Bout sound and vision

David Bowie

The papers in this issue are selected papers presented at the International Conference on Communication, VLSI and Signal Processing 2013, Siddaganga Institute of Technology, Tumkur, India.

Although the theme of the conference was signal processing, it is striking that almost all of the papers in the issue concern either sound or image analysis. One can postulate several reasons for this:

• Demand for electronic consumer products means that visual and audio sensors and interfaces are both highly developed and mass-produced. They are therefore both well-understood and inexpensive relative to their data collection power, making them viable for a wide range of practical applications.

• We as scientists are comfortable about using sensors where we can readily look at, or listen to, the data ourselves.

• Many applications revolve around a requirement to replicate human functions which, naturally enough, rely on human senses.

• Many artificial intelligence techniques are based on various models of human cognition. It makes sense to marry such models together with models of our own human senses.

The first of these arguments is certainly valid as far as image processing is concerned. A modern CCD is a miracle of miniaturized technology, and can be bought for just a few dollars. The argument is less convincing for sound. Microphones are still relatively expensive and bulky (at least if they are to exhibit good performance) and there are often significant additional costs for environmental shielding when using microphones in the field or for industrial applications.

The second argument is a tempting trap to fall into. But we should do well to remember the limitations of the human body. The human eye is only sensitive to a tiny window into the electromagnetic spectrum. And there are many phenomena, such as magnetic fields, which may be extremely useful when tackling an engineering problem but which the body cannot sense at all.

The third argument has its merits. For example, speech-to-text conversion is clearly a problem which falls into this category. More fundamentally, communication is seen by many to be one of the main pillars of intelligence, as epitomized by the Turing test. Thus there is a continued and ongoing interest in building intelligent systems which replicate aspects of human communications.

The final argument is much less well understood and an area of enormous research interest. The human brain is complex, with many structures which are clearly quite specific to their function, although as yet our understanding of how is often imperfect. What is clear though is that much information processing involves interplay behind several such structures and at different levels and often combining different inputs. Our current machine equivalents of these aspects are hybridization and data/sensor fusion. It is pleasing to see several such approaches within the papers of the issue.

Finally one must consider that the four issues I raise here cannot be considered only on an individual basis. There is an interaction and getting the balance right is one of the main challenges of the designer of any intelligent system which uses empirical real-world data.