The final thesis deals with different methods for classifying sound events of noise in the environment. Special emphasis is placed on the method of k-averages as one of the most suitable algorithms for processing non-phonetic and non rhythmic sound phenomena. The measurements performed under controlled conditions serve to support an analysis in which we focused on the role of the time constant. For this purpose, we have created program code in which the computer will learn independently on the basis of parts of the obtained data. The aim of the program is to determine the values of the characteristics for which the classification will be performed on the basis of the analysis of measurements, and to show the impact of changing the time constant as a tool for generalizing these values of the classification result. The measuring signals at the measuring point were captured by a microphone, and using an A / D converter, they were converted into a digital form suitable for computational analysis. After proper processing in the “Python” programming language, we were able to determine that the time constant strongly influences the data set used in machine learning and the perception quality results. To display the results, we used a so called confusion matrix that represents one of the methods for visualizing classifier performance.
|
