Abstract: According to the non-stationary characteristics of Chinese initials. A Chinese initials time-frequency structure analysis method based on the Chirp atoms Matching Pursuit（MP） decomposition is proposed. Simulation results show that compared with the direct Wigner-Ville Distribution（WVD）, the new method has no cross-terms；compared with the Gabor atoms decomposition, when the reconstruction initials signal energy reaches 90%, it can extract the initials time-frequency characteristic parameters more accurate, and the number of atoms used and time-consuming in the new method account for 30.77% and 24.53% of Gabor atoms decomposition respectively, confirming the superiority of the method.

Key words: Chinese initials, Matching Pursuit（MP）, Chirp atoms, time-frequency structure

摘要： 针对汉语声母语音呈现的非平稳特性，提出一种基于Chirp原子MP分解的汉语声母时频结构分析方法。实验仿真结果表明，该方法与直接WVD相比，无交叉项干扰，与Gabor原子分解相比，在重构声母信号90%能量情况下，不仅更加准确地提取了声母的时频特征参数，且采用原子个数和耗费时间分别为Gabor原子分解的30.77%和24.53%，证实了该方法的优越性。

关键词: 汉语声母, 匹配追逐（MP）, Chirp原子, 时频结构