An electrocardiogram (ECG) records the heart’s electrical activity, providing essential insights into cardiac health. Traditional visual analysis of ECGs is time-consuming and limits continuous monitoring, highlighting the need for automated signal processing. This paper introduces a new blind source separation method, cICABMGGMM, combining a bounded multivariate generalized Gaussian mixture model with constrained independent component analysis, addressing the typical independence assumption of ICA. We further propose the adaptive version, acICABMGGMM, which dynamically adjusts the association between estimated and reference sources using prior information without enforcing incorrect constraints. Our models are evaluated through simulation experiment and real-world applications in fetal ECG extraction and arrhythmia identification, demonstrating superior performance compared to state-of-the-art blind source separation methods.