Spectrogram Generation for Music Game Based on Attention-BiLSTM

Music game has been popular among digital entertainment since 1960s, and it is significant for children’s musical interests and the inculcation of music. The most important part of music game, charts, however are rarely created by professional staff. In many music games, charts are created by players themselves, resulting in uneven quality. The tedious and labor-intensive production of charts, and the negative impact it can have on the game experience, make them an important and difficult aspect of music game. Most research on music games has conducted from a sociological or design perspective, and there is little on automatic spectrogram for music games. Moreover, BiLSTM neural network mode has achieved remarkable success in processing temporal problems such as speech recognition. Meanwhile, Attention excels in enabling models the ability to discriminate. In this paper, we proposed an automatic generation model combined Attention mechanism with BiLSTM neural network for music game.