It is known that many real signals, such as speech signals, are non-stationary processes that express theirfeature through changes in parameters over time. By observing speech in shorter time intervals, the property ofstationarity can be notice. This characteristic enables the application of techniques for adaptation to local signalcharacteristics in signal processing algorithms. Many of these algorithms are described by standards, and in additionto intensive development in this area in last decades, there is a constant need for the development of new solutionsand standards. One of the most used parameters of the speech signal for adaptation is the mean (average) signalpower. The change of speech in time results in a wide dynamic range of average power. In addition to the predicteddynamics of average power, in the design of systems with adaptive techniques it is important to include otherparameters, among which is the Probability Density Function (PDF) of the average power. The goal of the researchpresented in this paper is the analysis of the probability distribution of the average power of speech signals, based onwhich the adaptability in the development of algorithms in digital processing would be improved, which wouldensure higher quality and less requirements in data transmission and storage. In addition to the theoreticalconsideration, the research was conducted on real speech signals of different speakers. In modern technical systems,where Internet technologies are prominent, processing, transmission and memorization of speech is executed frameby frame. Therefore, an analysis of the probability density of the average power for different frame lengths wascarried out in the experiment. In the experimental part, for each of the speech signals of the speech corpus, theProbability Density Function that best describes the average power values per frame was determined. Experimentalresearch results indicate that the function that best describes the average power is different for different speakers. Inaddition, when observing one speaker, the Probability Density Function is different for different frame lengths. Itcan be concluded that when it comes to adaptive techniques in the digital processing of the speech signal, it isimportant to consider the characteristics of the average power, among which is the Probability Density Function ofthe average power
Read full abstract