Recognition and Error Correction Techniques for Piano Playing Music Based on Convolutional Cyclic Hashing Method

Music as a sound symbol can express what people think; music is both form of social behavior and promote people’s emotional communication; also entertainment enrich spiritual life. In this paper, we propose new convolutional recurrent hashing method CRNNH, which uses multilayer RNN to learn discriminate piano playing using feature map sequences. Firstly, sequence learning preserving similarity hash function designed consisting maps extracted from multiple layers pretrained CNN; secondly, deep framework proposed generate codes RNN, directly the input preserve spatial structure maps; finally, loss semantic balance codes, while considering quantization error generated when layer outputs binary codes. The experimental results illustrate that CRNNH obtain better performance compared other methods.