The initial part of a Chinese syllable is usually known as carrying less energy and more frequency components in a scattered way while the final part carrying much more energy as well as less frequency components which are focused in a lower frequency range when compared with each other. By employing time-frequency analysis, these differences were represented with corresponding atoms' characteristics in time-frequency plane. With the genetic algorithm be introduced to reduce the matching pursuit computing load for searching best atoms, a new segment method based on Gabor atomic parameters was implemented. Simulation on 115 syllables has achieved an accuracy of 80.87%.