<p>Watermarking supports inserting and hiding secret data with predetermined host signals, such as audio, video, and images, without interfering with regular use. Digital audio watermarking is used to embed the images within the audio signal. Digital audio watermarking poses unique challenges due to the sharp sensitivity of the human auditory system compared to visual systems. Conventional methods exhibited several limitations such as limited transparency, robustness, additive noise, and increased complexity. Due to this increased computational complexity, the conventional approach cannot be used in real-world applications. Therefore, this research proposed a Hybrid Adaptive Amplitude Modulation and Optiphase Modulation (HA2M-OpM) technique, which effectively embeds secret messages into the audio signals. The proposed ultra-search optimization algorithm is designed using the historical Adaptive Moment Estimation (Historical Adam) optimizer with the DNN, which ensures effective convergence and robustness in the block selection process. Ultra-search optimization is used to identify the best blocks for secret message insertion. The adaptive learning rates of the search optimizer prevent overfitting which enables the proposed model to generalize across diverse audio samples. Additionally, the three-level decomposition technique used in this model enhances the audio quality. The comparative evaluations demonstrate that the HA2M-OpM model outperforms existing techniques with 48.15&#xa0;dB of Peak Signal-to-Noise ratio (PSNR) for the Flower Color Image dataset, making it highly applicable in real-world scenarios.</p>