Feature Recognition Method of Digital Piano Audio Signal Based on CNN

Among the various methods of cultivating students’ music feeling rhythm, listening, reading and moving require teachers to play music on CDs and tapes or play relevant music on the piano. As the “king of musical instruments”, piano has rich timbre and shocking expressiveness, which is deeply loved by music lovers. The performance of piano is not only restricted by the technical level of performance, but also restricted by people’s mental and physical state. People may play wrong notes or play inaccurate rhythm when they are in poor mood or feel slightly tired after playing for a long time; The tone of the piano is single, and so on. This paper studies the feature recognition of digital piano audio signal based on CNN, extracts the correlation spectral feature of digital piano audio signal, then constructs the spectral feature decomposition model of digital piano audio signal by using wavelet packet transform method, extracts the detailed feature of digital piano audio signal, and according to the time-frequency feature distribution and spectral feature clustering of digital piano audio signal, Realize the feature optimization recognition of digital piano audio signal. The computer recognition system of piano playing music has preliminarily realized the function of automatic recognition of the machine. It is hoped that the automatic recognition can be brought into good play in the fields of auxiliary music education.