This article is devoted to solving the problem of recognizing various sounds in the environment, which is widely used in surveillance and control systems and allows you to identify objects of various nature, for example, a car, a boat, an airplane, animals, birds, etc. The paper proposes an algorithm for recognizing sounds in an audio signal based on the analysis of the signal frequency components corresponding to the discrete cosine transform coefficients of signal fragments. The discrete cosine transform provides, in contrast to the Fourier transform, the decomposition of the signal into real frequency components, which reduces computational costs when implementing the algorithm. In the developed algorithm, based on the frequency analysis of the audio signal, as an example, notes of different octaves are determined. At the stage of preprocessing, fragments corresponding to pauses are allocated in the initial signal and the informative audio signal fragments are formed, during the analysis of which, at the next stage of the algorithm, notes are recognized. Computational experiments with a model sound signal demonstrated the developed algorithm.