Currently, music recognition research is primarily focused on single note recognition, with some limitations in recognition accuracy and antinoise performance. This paper proposes a new algorithm for piano playing music recognition against the backdrop of intelligent interaction. The method of spectrum peak sorting is extended to the field of multifundamental frequency detection, and high and low channel processing is achieved. The statistical properties of spectral entropy of coefficients in compressed domain are used, resulting in more stable fingerprints. This statistical feature will not be destroyed after the original segment is processed, ensuring that the calculated feature maintains its high stability. This method can effectively improve the accuracy of fundamental frequency extraction by highlighting the peak characteristics of the periodic position of frame samples, avoiding the influence of half-frequency and frequency doubling, and thus avoiding the influence of half-frequency and frequency doubling. In comparison to traditional methods, achieve higher accuracy and fault tolerance. The feasibility and efficiency of the algorithm proposed in this paper are confirmed by a simulation experiment. This method’s overall performance meets certain practical requirements and achieves the expected results, laying the groundwork for future research in this field.