Frame change ratio:A measure to model short-time stationarity of speech

In many applications such as speech coding and PLC (Packet Loss Concealment) methods, short-time stationarity of speech is assumed. In this paper we introduce a new measure to model the short-time stationarity of speech. The model is developed by using a large amount of speech data together with real speech codecs and PLC methods. Experiments show that a single parameter, the frame change ratio, can be used to model this characteristic. This parameter is independent of speech material, speaker gender, and loss rate. It provides a generic characteristics of the deviation of speech from stationary behavior when evolving from one frame to the next.