Wearable devices acquiring physiological signals like Photoplethysmogram (PPG) are prone to motion artifacts, necessitating effective denoising to ensure reliable cardiovascular data extraction. Traditional denoising methods risk degrading the underlying signal, and while AI-driven approaches, such as deep learning (DL), show promise, they struggle with random and systematic distortions and often require large datasets for accurate model training. To address these challenges, this paper introduced a style transfer–assisted cycle-consistent generative adversarial network (stccGAN) to denoise a 3-channel PPG (red, green, and infrared wavelengths) signal acquired by the Soundi chest sensor (patented and certified). The study employed two identical devices: one gathering the chest PPG signal (to be denoised), and the other synchronously acquiring the finger PPG signal (reference signal). The method employs style transfer DL, utilizing a dual-generator architecture with U-Net, GRU, and LSTM layers to process the multidimensional signal and enhance the chest PPG quality. The system was validated with data from 30 subjects, 20 used for training and 10 for testing. The results showed over 70% correlation with the reference signal on average, with an improvement of about 15% with respect to the acquired chest PPG. This indicated promising restoration suitable for cardiac assessment and blood pressure estimation. The proposed method was able to restore a good PPG quality making it usable to perform evaluations of the cardiac condition and estimate the blood pressure.