ISCA Archive SUS 1995
ISCA Archive SUS 1995

Discussion on "emotion"

Claude Legros

In some cases, emotion leads to extremely unpleasant situations, and even tight ones. It is the case of aeronautic failures and all other situations dealing with security of people.

Therefore, one has always been interested in knowing the real psychological (emotional) state of a person. But the word "emotion" covers a very wide range of definitions corresponding to as many as states. So, before going on, we must answer the following preliminary questions: "What emotion are we studying?" and "Why are we studying emotion?" Two reasons arise: the simple knowledge and the quick identification of the emotion in order to act to try to cancel it and its possible effects.

One can quote some examples in which the early knowledge of the psychophysical state could help efficiently to clear up a rising difficulty: aeronautic small incident which could deteriorate into severe one, admission service of a bank, heavy trucks drivers, and so on...

These are one of the reasons why a lot of researchers from all domains asked themselves on the way to follow to estimate the generally said "emotional state" of a person.

Previous to any estimation of emotion, it is needed to define it. Scherer, in its 1986 report, said that "a greater precision in the description of the emotional state to be studied" is needed to go further. How to Qualify the Emotional State of a Person?

The first idea is to ask the person concerned on his (her) own feeling. It has already been said (Scherer 1986) that the answers are not accurate enough to be reliable all the more because these estimations, for a lot of reasons, are often unclear. Nevertheless, emotions can be divided into several usual categories, but acoustical correlates are not so easy to perform.

As this direct investigation is not really usable, it remains some indirect features: emotion could be characterized by its effects on people submitted to it.

One of the ways followed is based on medical analyses. It seems that they are quiet reliable in some cases such as stress. But, estimation is very huge and needs a great amount of sensors. Furthermore, it seems that other kinds of emotions, such as anxiety, fear... are much more difficult to investigate by these ways.

Finally, it soon appeared that the emotion state acts on the speech of the person and already during the 40's researches arised. The main benefit of speech investigations is that the data are very easy to collect. It is now well known that speech quality changes under the effect of emotion. But these changes vary with the kind of emotion and the same parameter may vary in opposite way for two different emotion types.

Voice parameters extracted from speech are numerous but could be divided into 3 main categories: fundamental frequency and derivatives, spectral investigations (specially formants analyses) and duration criteria.

A very large amount of researches have been conducted on these topics and as the correlates between emotions and acoustics are now obvious, it is therefore needed to consider the problem as a whole to be able to go further. Simulated Versus Real Situations

Most of the studies are performed with actors simulating the different emotions. This is not a real situation and the comparison between the two situations may become rather difficult because the psychological state is not at all the same. By the way, some simulated situations can reach the real ones (e.g. flight simulators for pilots training).

Therefore, let us point out the main questions we have to answer, keeping in mind that we are rather interested in stress. Main Questions What is emotion? What is the best way to define it? Has stress the same stutus as other emotions? What kind of emotion is speech able to deal with? Is it possible to graduate emotion in order to follow to the closest its growth and, may be, to act before the situation becomes serious? Is it needed to graduate emotion by a separate way from acoustics (psychology, medicine...) and then to graduate the acoustic parameters, or is it better to directly graduate emotion by the way of the fluctuations of the acoustic parameters? Are laboratory experiments a good way to estimate what could happen in real cases, and how to skip from one to the other?

All these questions shall be discussed during the workshop.