The performance of an automatic speech recognizer degrades when there exist
s an acoustic mismatch between the training and the testing conditions in t
he data. Though it is certain that the mismatch is nonlinear, its exact for
m is unknown. Tackling the problem of nonlinear mismatches is a difficult t
ask that has not been adequately addressed before. In this paper, we develo
p an approach that uses nonlinear transformations in the stochastic matchin
g framework to compensate for acoustic mismatches, The functional form of t
he nonlinear transformation is modeled by neural networks. We develop a new
technique to train neural networks using the generalized EM algorithm. Thi
s technique eliminates the need for stereo databases, which are difficult t
o obtain in practical applications. The new technique is data-driven and he
nce can be used under a wide variety of conditions without a priori knowled
ge of the environment, Using this technique, we show that we can provide im
provement under various types of acoustic mismatch; in some cases a 72% red
uction in word error rate is achieved.