A multimodal approach to extract optimized audio features for speaker detection | IEEE Conference Publication | IEEE Xplore