Extraction of Audio Features Specific to Speech Production for Multimodal Speaker Detection