Piano Music Teaching under the Background of Artificial Intelligence

The work is aimed at solving the problems of easy trapping into local extremes and slow convergence speed traditional music teaching evaluation system on Backpropagation Neural Network (BPNN). note recognition methods are susceptible to high noise complexity. Firstly, Levenberg Marquardt (LM) algorithm used optimize BPNN; secondly, an improved endpoint detection proposed by short-term energy difference, which can accurately identify time value each in piano playing audio. By frequency domain analysis method, a radical extraction standard harmonic note’s pitch. Finally, performance model BPNN implemented, implemented Musical Instrument Digital Interface (MIDI) system. This be correct errors students’ performances process perform overall evaluation, rhythm expressive evaluation. Teachers students play minuet collect experimental samples train test model. practical result shows that (1) after 3000 times training, neural network error less than 0.01, converges; (2) results designed basically line with actual level performer have specific feasibility; (3) optimized during accuracy rate 94.3%, 5.25% higher method. correction for pitch 92.9%, 5.21% has significantly notes pitches played player. effectively help beginners improve efficiency practice. purpose this study alleviate scarcity teachers, reduce intensity realize automatic objective playing, provide necessary technical support improving teaching.