Extracting deep bottleneck features using stacked auto-encoders | IEEE Conference Publication | IEEE Xplore