Hybrid Audio Watermarking Using Frequency-Time Features and Machine Learning for Enhanced Robustness Against Attacks

This paper offers a strong and imperceptible audio watermarking system leveraging frequency-time domain capabilities and machine learning to know classifiers to embed and extract text-based total watermarks. The model integrates Support Vector Machine (SVM), Random Forest (RF), and K Nearest Neighbor (KNN) classifiers to assess the system's performance underneath commonplace attack scenarios, together with noise addition, MP3 compression, and filtering. Key output parameters along with Bit Error Rate (BER), Signal-to-Noise Ratio (SNR), Accuracy, and Normalized Cross-Correlation (NCC) have been analyzed to evaluate robustness and constancy. Results imply that SVM consistently outperforms RF and KNN, accomplishing the lowest BER (2.5%), maximum SNR (29.3 dB), and superior accuracy (98.0%). RF furnished slight robustness, at the same time as KNN demonstrated decreased resilience to attacks. Comparative evaluation and MATLAB-based total visualizations highlight SVM's effectiveness in maintaining watermark integrity and imperceptibility. This look establishes device learning as a pivotal tool in designing audio watermarking structures resilient to common distortions. Future work objectives are to discover hybrid models and extend the assessment to diverse real-world datasets for further validation.