A deep generative speech denoising model based on Transformers and the U-Net architecture