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 generic characteristics of the deviation of speech from stationary behavior when evolving from one frame to the next