Noise adaptive training using a vector taylor series approach for noise robust automatic speech recognition.