The study examines the correlation between driver behavior and speech emotion, proposing a method for recognizing four driver behavior states: talking on the phone, laughing, sleepy, and normal driving. It employs a silence removal technique using short-term energy and zero crossing rate to enhance feature extraction via the Mel frequency cepstral coefficient and utilizes a multi-layer perceptron classifier. Experimental results indicate an accuracy range of 58.7% to 76.6%, suggesting potential for a driver behavior identification system aimed at improving road safety.