Improving robustness of connectionist speech recognition systems bygenetic algorithms

摘要

We present an approach which limits significantly the drop ofperformances related to automatic speech recognition systems (ASRSs)caused by acoustic environment changes. We propose to combine principalcomponent analysis (PCA) and genetic algorithms (GA) in order totransform the noisy acoustic environment into a predefined andwell-known (canonical) environment. The idea consists in projecting thenoisy speech parameters onto the optimal subspace generated by thegenetically modified principal components of the canonical environment.The results show that in noisy and changing environments, the proposedPCA/GA optimized system achieves high recognition rate compared to thebaseline system