Feature generation based on maximum normalized acoustic likelihood for improved speech recognition | IEEE Conference Publication | IEEE Xplore