This paper proposes a new approach for automatic syllable segmentation of Mandarin spontaneous speech. Automatic speech segmentation is important for continuous speech recognition because it reduces the search space effectively in automatic speech recognition. Moreover, the signal segmentation technique is useful in automatic speech marks and labels. However, for automatic speech recognition (ASR), it is difficult to segment the speech input reliably into useful sub-units because (1) syllable units can often be located roughly via intensity changes, but exact boundary positions are elusive in successive vowels, (2) energy changes in speech spectrum or amplitude help to estimate unit boundaries, but these cues are often unreliable due to co-articulation, and (3) finding boundaries for units bigger than phonemes combines the difficulties of detecting phoneme edges and of deciding which phonemes group to form the bigger units. In this paper, we present a hybrid segmentation method that utilizes silence detection, convex hull energy analysis, and spectral variation analysis. Furthermore, Hamming short-time sliding-windows were applied twice on audio signals to get more obvious convex hull valleys. Mandarin speech segmentation was used as a testing case, and the effectiveness of the proposed segmentation system was confirmed by the experimental results.