Speaker Diarization Based on Speech Signal Approximation by Step Function

In this paper, we describe a new method for speaker clustering in an audio file. The main idea is to replace the speech signal with a step function having a limited number of levels. The research goal is to determine the signal characteristics obtained from the analysis of the step function produced. The step function is created by setting multiple levels that divide the signal range into non-overlapping strips. All the source signal values, which are inside a stip, are changed for the strips mark. Using the sine function as a template, we get recommendations for choosing the sources best-keeping features. We employ the obtained results to solve the problem of speaker diarization. The developed diarization algorithm requires little computer resources. The experiments show the suitability of the developed results to the conditions of the online diarization process.