Abstract: For overweight tail characteristics of the speech signal, this paper proposes a mixed-laplace distribution acoustic model. Laplace mixture distribution model parameter estimation is discussed theoretically, and which can be achieved from the principles and algorithms. The maximum expected（Expectation Maximization, EM）algorithm has been applied to obtaining good results. The hybrid Laplace model  has been creatively used to study speech signal processing.

Key words: mixture Laplace model, speech signal, Expectation Maximization（EM）, kurtosis, heavy-tailed distribution

摘要： 针对具有超重尾特性的语音信号，提出了混合拉普拉斯分布语音模型。从理论上探讨了混合拉普拉斯分布模型的参数估计，从原理与算法得以实现。通过最大期望（Expectation Maximization，EM）算法取得了良好效果。创新运用混合拉普拉斯模型研究语音信号处理。

关键词: 混合Laplace模型, 语音信号, 最大期望算法, 峰度, 重尾分布