Automation of the audio mastering process using machine learning algorithms

Abstract

The article examines the problem of automating the audio mastering process using machine lear-ning algorithms. The relevance of the study is determined by the fact that modern digital tools have made music production accessible to a wide range of users, while high-quality mastering still requires specialized knowledge, critical listening skills, technical control, and significant time resources. This creates a barrier for independent musicians, small studios, and creators who do not have access to professional audio engineers. The paper summarizes the main stages of audio signal analysis, including spectral analysis, equalization, dynamic processing, loudness control, and the evaluation of signal properties using objective and subjective criteria. Particular attention is paid to the capabilities of convolutional neural networks, recurrent neural networks, and LSTM networks in audio processing tasks, music genre classification, and the analysis of the temporal structure of an audio signal. The article analyzes scientific approaches to sound quality assessment, automatic mastering, and the use of reference or professionally mastered audio recordings as a basis for training models. The GTZAN dataset, which contains music fragments of different genres, is considered as an example of an experimental basis. A comparison of neural network types is presented in terms of classification accuracy, number of training epochs, and approximate training time. The summarized results indicate that CNNs are appropriate for spectrogram analysis and frequency feature extraction, while RNNs and LSTMs are better suited for sequential data and musical context modeling. The study concludes that machine learning can serve as an effective foundation for tools supporting automated mastering; however, complete replacement of human expertise remains premature due to the importance of artistic intention, genre context, and subjective listener perception.

Keywords: machine learning, mastering, sound recording, artificial intelligence, audio processing, sound design, neural networks.