In today’s rapidly developing technology, achieving autonomous performance of robots through machine learning has become a highly challenging aspect of music technology. Therefore, this article intends to construct a novel CMPN cross modal aggregation network and apply it to the intelligent transformation between piano music and performance. This article strengthened the extraction of important information on rhythm and beat in music works through attention mechanisms. On this basis, a machine learning based piano performance method was adopted. In the exploration of cross modal fusion effects, under the experimental conditions of visual and auditory fusion, the accuracy of note recognition was 92%, and the accuracy of rhythm grasp was 88%. The CMPN cross modal aggregation network in this article can not only promote the development of music technology, but also open up new avenues for the development of music teaching and performing arts, and lay a good foundation for future intelligent music performance systems.