Image, sound and video are three different embedded requirements. In terms of products, image\/video applications are mostly based on arm platform (Huawei's hisilicon for video processing, TI's DMx series, most of which are arm's core chips), while dsp is the mainstream for sound processing. cuda and FPGA are rarely used in actual embedded products, but may be necessary platforms in the development and design process.
cuda is mainly used for training models in image recognition based on deep learning, while FPGA is mainly used for algorithm engine verification platform in the design of related IC chips, and rarely directly used in the final embedded products, because the same performance resources of FPGA are too expensive compared with arm and dsp, and the same is true for graphics cards.