The Blind Source Separation (BSS) is crucial in processing the audio signals to separate the independent signal sources from the mixed signals. However, there are some challenges when applying the existing methods in real-world scenarios to accurately and efficiently identify mixed signals. For a better real-time implementation of BSS, a Dynamic Spectral Independent Component Analysis (DS-ICA) is proposed in this research, which involves incorporation of dynamic spectrum for temporal correlation and blend spectral analysis for source signals. Compared to the conventional approaches, DS-ICA makes use of another model that employs both cubic splines and indications for functions estimation of spectral density and mixing matrices. Moreover, DS-ICA also uses Copulas to capture and quantify the dependencies between the source signals, thus it is a more effective method to isolate the sources that are mixed together dependently. Since DS-ICA is a novel method, it has been comprehensively tested and its benefits have been explained and proved through simulation experiments and real-world applications such as cockpit voice signal separation, transformer acoustic signals separation, and bioacoustics signals separation. The proposed DS-ICA achieved better accuracy with 98.51%, and Signal-to-Noise-Ratio (SNR) of 6.85 dB, which provided relatively good solution for improving BSS, when compared to existing methods FastICA and Convolutive Mixtures based ICA (CM-ICA).