本文为德国慕尼黑技术大学（作者：Konstantin Lackner）的学士论文，共56页。

近几年来，人工智能（AI）的研究不断取得进展，主要是因为在人们的数字生活中，基本上每个部分都会有大量的数据产生，因此，AI算法可以得到非常密集且准确的训练。

In the recent years the research onartificial intelligence (AI) has been increasingly progressing, mainly becauseof the huge amounts of generated data in basically every part of one’s digitallife, out of which AI algorithms can be trained very intensively andaccurately.

除此之外，现代硬件计算能力的进步也促进了这一领域的蓬勃发展。

Beyond that, the progress in computationcapabilities of modern hardware helped this field to flourish.

到目前为止，已经开发出一些可以超越人类能力的人工智能方法，例如国际象棋计算机DeepBlue或IBM的Watson，他们在“Jeopardy”游戏中击败了人类的顶级高手。

So far, certain methods to implementartificial intelligence have been developed that could outperform human abilities,such as the Chess Computer DeepBlue or IBM’s Watson who beat the best humans inthe game “Jeopardy”.

实现人工智能的一种方法是人工神经网络，它是由人类或动物大脑的工作方式驱动的。

One method of implementing ArtificialIntelligence are Artificial Neural Networks, which have been developed motivatedby how a human or animal brain works.

人工神经网络越来越成功地完成了模式识别等任务，例如语音和图像处理。

Artificial Neural Networks haveincreasingly succeeded in tasks such as pattern recognition, e.g. in speech andimage processing.

然而，在音乐创作等创造性工作中，这方面的研究却很少。

However, when it comes to creative taskssuch as music composition, only little research has been done in this area.

本论文主要研究人工神经网络的音乐创作能力。

The subject of this thesis is toinvestigate the capabilities of an Artificial Neural Network to compose music.

特别地，本文将着重于一个特定的和弦序列的旋律组成。

In particular, this thesis will focus onthe composition of a melody to a given chord sequence.

本文主要目标是实现一个长短期记忆（LSTM）的递归神经网络（RNN），该网络实现的旋律听起来很悦耳，无法与人类的旋律区分开来。

The main goal is to implement a long-shortterm memory (LSTM) Recurrent Neural Network (RNN), that composes melodies thatsound pleasantly to the listener and cannot be distinguished from humanmelodies.

此外，对作曲旋律的评价也起着重要作用，以便能够客观评价LSTM RNN作曲的质量，以进一步对该领域的研究做出贡献。

Furthermore, the evaluation of the composedmelodies plays an important role, in order to objectively asses the quality ofthe LSTM RNN composer and therefore be able to make a contribution to theresearch in this area.

1 引言

2 算法合成的最新进展

3 神经网络

4 数据表示形式：MIDI

5 具体实现

6 实验

7 评估

8 结论

附录A 测试数据得分

附录B 网络构成得分

附录C 人声乐谱得分

更多精彩文章请关注公众号：