A　new　driver　monitoring　system　called　DeriskNet　for　driving　risk　prevention　is　proposed　in　this　paper.　We　first　develop　a　deep　convolutional　neural　network　to　recognize　the　driver＇s　emotions.　Then,　we　devise　an　audio　on　demand　mechanism　to　automatically　collect　audio　resources　via　web　crawling　for　preventing　driving　risks　from　driver＇s　negative　emotions.　Experiment　results　show　that　our　DeriskNet　provides　superior　accuracy　and　reliability　than　the　conventional　convolutional　neural　network　designs　in　terms　of　facial　emotion　recognition.　©　2018　IEEE.