The effectiveness of model-based Bayesian systems was proved in the area of ECG analysis. In this article, we offer a novel Bayesian framework for ECG analysis which makes use of the Kalman filter bank as well as the expectation maximization (EM) method. In comparison with earlier Bayesian algorithms, this approach simply needs an R-peak identification phase during the ECG analysis. Every ECG pulse can be divided into two parts, namely the QRS complexity (high-frequency segment) and the P&T waves (low-frequency portion), based on its position of R-peaks. The low and high frequencies are denoised using the Kalman filter bank for the above process, which consists of two separate Kalman filters. The expectation maximization (EM) method is used to calculate and repeatedly modify the settings for each of these filtration. A number of ECG databases containing patterns with morphological deviations and changes, such as atrial premature complexes (APC) and premature ventricular contractions (PVC), were used to assess the suggested technique. A number of ECG blurring techniques, including wavelet change, empirical mode breakdown, and band-pass filtering, have been compared to the suggested approach. The outcomes demonstrated that the suggested technique performed superior to reference techniques at small input SNRs from both SNR enhancement and multi-scale entropy-based weighted distortion (MSEWPRD) views.