ISCA Archive SUS 1995
ISCA Archive SUS 1995

How do people cope with stress in the use of speech recognition?

Chris Baber, Claude Legros

Aims of Discussion: to consider the different strategies that people adopt in stress conditions; to relate the experience of discussants in dealing with stress conditions; to outline possible solutions to the problems of stress, particularly in terms of the interaction between users and speech recognition devices.

Outline of key topics

It is clear that various forms of stress can have an adverse affect on the success of speech recognition systems. The workshop has considered a number of different forms of stress, e.g., environmental stress such as noise, emotional stress, workload stress. Each of the different types of stress appears to share some general effects and to have some effects peculiar to the form of stress. The effects can be considered in terms of the consequences for speech recognition performance, the consequences for human-computer interaction, and the consequences for human performance. For instance, consider vibration as a form of stress: the performance of the recogniser might be impaired by the mechanical effects of vibration, the human-computer interaction might be impaired by the instability of the speaker's voice, in terms of human performance control of motor and visual response might be impaired. In this instance, the problem of coping with stress would appear to be less one of adapting human behaviour (as the effects of stress would appear to lie outside the person's control) and more in one of adapting the technology. On the other hand, the effects of some forms of stress do have the potential to be modified by the person, e.g., in terms of noise, the effects on human performance can be overcome by adoption of an appropriate strategy, while the effects of human-computer interaction can be approached using appropriate algorithms to deal with phenomena such as the Lombard effect. This implies that people are, within some limits, able to modify the effects of stress on their performance and on their production of speech, and that it may be possible for system design to capitalise on predictable relationships between stress, performance and speech recognition.

People attending this discussion are asked to bring with them examples of applications and work that they have been involved in or know about in which stress has had an impact on both speech recognition performance and human performance. From these examples, the discussion will seek to develop possible causes of the problems and possible solutions for future developments.