I. Introduction
The widespread use of mobile telephones has motivated the development of robust speech recognition systems in cars [1]. A major source of errors in automatic speech recognition systems is the inaccurate detection of the beginning and ending boundaries. In cars, the problem is further complicated by nonstationary backgrounds where there may exist concurrent noises due to movements, engine running, speed change, braking, slams, etc. These background noises can be broadly classified into three classes: impulse noise, fixed-level noise, and variable-level noise. Decreasing the distance between the mouth and microphone is one way of minimizing the effects of such transient background noise. However, this method is not user-friendly. In order to solve this problem, many researchers proposed robust word boundary detection algorithms in the presence of noise. However, they focused only on the impulse noise and fixed-level background noise. The main aim of this paper is to develop a new robust word boundary detection algorithm to attack the problem of variable-level background noise in cars.