A heart sound signal (HSS) is sensitive to physiological noise and environmental noise, thereby degrading their quality, which makes the accurate diagnosis of machines or doctors difficult and unreliable. To this end, we present a heart sound denoising method using Parameterless Scale-space Boundary Detection (PSBD)-Empirical Wavelet Transform (EWT) and Enhancement Generative Adversarial Network (EGAN) to remove noises that corrupt HSSs in this paper. First, it introduces PSBD and kurtosis to find boundaries delimiting consecutive EWT modes. And then, it further selects the relevant modes on the Pearson's correlation coefficient between each of empirical modes and the original signal to reconstruct HSSs. Finally, EGAN is proposed to improve PSBD-EWT's generalization capacity with regard to different noises. Experimental validation is carried out on PASCAL, MHSDB and WUPHSD databases. The results show that our proposed method achieves significant improvements over state-of-the-art methods. In the case of white Gaussian noise with Signal Noise Ratio (SNR)=5dB, it obtains the best denoising performance under a SNR of 12.53dB and an Root Mean Square Error (RMSE) of 0.034.