It is a relatively complicated project that computers recognize emotions in human speech. Labeling the emotions and analyzing speech emotions visually in a voice conversation is a good way to approach this process. So, my research is aiming to label sentiment and find the reason that the accuracy is not good enough using relative intensity of emotions based on probability distribution. In particular, I will make an audio player for giving the corresponding emotion tags with time since speech is played. As an experiment result, the changes of emotion can be labelled on audio conversation in real-time efficiently.