Speech Translation with Foundation Models and Optimal Transport: UPC at IWSLT23