This dissertation adopts ASR based on Gaussian Mixture Model (GMM), so re-search emphasizes on robust features and speech enhancement techniques.

• 消除噪声所带来的失配可以映射到信号空间、特征空间和模型空间。