There are two common classes of audio, speech and music, in broadcast. An algorithm of segment and classification of audio designed in this paper will be applied to audio equalizer in the digital audio broadcasting transmitter. The algorithm described is based on zero crossing rate and energy that are able to separated non-silence segments from audio, and on Modified low energy ratio (MLER) and maximum likelihood criterion, which provides the ability to distinguish the two classes, speech and music. Modified low energy ratio is obtained on the basis of energy resulting in low computational complexity, leading to runs easily in real time. Experiment results to data show performance with the accuracy rate of over 96%.