MODERN METHODS AND PROBLEMS OF DEEP DIGITAL WATERMARKING OF AUDIO CONTENT

Abstract

Introduction. Digitalization of information has greatly simplified the process of creating, distributing, and sharing audio content. However, the increased availability of digital media has led to an increase in unauthorized copying and distribution of audio files, which violates intellectual property rights. Currently, various methods are used to protect audio content, including file encryption, digital rights management (DRM), digital watermarking technologies, authentication methods, etc. Digital watermarks stand out among these technologies due to their ability to provide content tracking and allow for the identification of file origin. These methods helps you to identify the owner and protect the content even when it is further distributed. However, despite the achievements, a number of unresolved issues remains. The most significant among them are: ensuring the resistance of watermarks to various types of audio processing and high-quality hiding of watermarks, which should be invisible to listeners and preserve sound quality. The need for more reliable and versatile content protection methods remains a hot topic for research and development today. Purpose. To investigate the current state of the digital watermarking of audio content, methods and technologies of the process of deep digital watermarking of audio content, to assess their level of quality, security and to outline promising areas of research in the field of digital watermarking of audio content. Methods. To achieve this goal, a combination of qualitative and quantitative research methods was used. Namely, scientific research methods, in particular, a systematic review and analysis of scientific articles, patents, industry reports on practical implementations of existing digital watermarking methods, methods for analyzing the security and resilience of watermarking methods, in particular, the DCT-MLP-LSB method, benchmarking tools (AudioMarkBench) to conduct a comparative analysis of the resilience and invisibility of watermarking methods based on their overall effectiveness, a method of comparing results and a method of synthesizing results. Results. Based on the conducted research, a comprehensive understanding of the current state of digital audio content marking is obtained, the main areas of application are considered, and promising areas of application of these technologies are highlighted. In particular, 7 main segments of digital watermarking for audio content have been identified: - authentication; - copyright; - broadcast verification; - audience measurement; - expertise; - data hiding; - communication. An analysis of the literature shows that by 2020, 80% of research in this area was conducted to watermark confidential information, and only 20% to process standard content. Today, research in these two segments is evenly distributed, and parity has almost been achieved. The rapid development of artificial intelligence algorithms causes a sharp increase in interest in watermarking standard content. The key aspects of the two main processes (embedding/extracting) of the watermarking system are investigated and schematized. The 4 key parameters that have the most significant impact on the performance of audio content watermarking systems are identified and considered: - security; - capacity; - reliability; - invisibility. A study was conducted to evaluate the effectiveness of modern methods of digital audio content watermarking and identify gaps in existing research. Despite numerous studies in automatic sound classification, modern methods only partially reproduce human perception. As a result, these methods work well in some cases, but are practically unusable in others. We propose to replace the MLP (Multi-Layer Perceptron) model with the SVM (Support Vector Machine) model in the pipeline of the hybrid DCT-MLP-LSB algorithm in order to develop it. The main directions of promising research in the field of digital watermarking of audio content are identified. Conclusion. The conclusions of the article demonstrate the relevance of the research topic and outline a segment of unresolved problems in the field of audio content watermarking. Keywords: digital watermark, audio file, security, methods of audio content protection, digital watermark technologies, deep learning, neural network.