This chapter discusses statistical speech processing methods. It deals with statistical parameters such as mean, standard deviation, skew, and kurtosis. The speech recognition process consists of matching the incoming speech template with the stored templates. The template with the lowest Euclidean distance from the input pattern is the recognized word. When the two sequences of feature templates of two different speech words are to be compared, the sequences must be warped dynamically. Statistical framework allows density estimation, alignment of training data, and silence detection for isolated or continuous word recognition. Speech recognition methods use statistical modeling of speech such as the hidden Markov modeling (HMM) and Gaussian mixture models. HMMs for speech recognition are the group of states interconnected. The states are assumed to emit feature vectors depending on the probability specified for each transition.