An automatic music generation and evaluation method based on transfer learning

Abstract

In recent years, deep learning has seen remarkable progress in many fields, especially with many excellent pre-training models emerged in Natural Language Processing(NLP). However, these pre-training models can not be used directly in music generation tasks due to the different representations between music symbols and text. Compared with the traditional presentation method of music melody that only includes the pitch relationship between single notes, the text-like representation method proposed in this paper contains more melody information, including pitch, rhythm and pauses, which expresses the melody in a form similar to text and makes it possible to use existing pre-training models in symbolic melody generation. In this paper, based on the generative pre-training-2(GPT-2) text generation model and transfer learning we propose MT-GPT-2(music textual GPT-2) model that is used in music melody generation. Then, a symbolic music evaluation method(MEM) is proposed through the combination of mathematical statistics, music theory knowledge and signal processing methods, which is more objective than the manual evaluation method. Based on this evaluation method and music theories, the music generation model in this paper are compared with other models (such as long short-term memory (LSTM) model,Leak-GAN model and Music SketchNet). The results show that the melody generated by the proposed model is closer to real music.

Technical, Musical, and Legal Aspects of an AI-Aided Algorithmic Music Production System

“…[31] is devoted to measuring and evaluating musical excerpts of classical piano music generated by deep learning models based on the results of the blind test conducted on a group of musicians and non-musicians. In turn, in [32], an evaluation method for music (generated by deep learning) was proposed as a combination of mathematical statistics and music theory evaluation. However, this approach is not possible in our case due to the complexity of composed music.…”

“…Claimed objectivity of such methods is based on arbitrary selection of rules, which in natural music changes over history and varies between styles. They may evaluate similarity between fragments of melody and rhythm [32] but not their objective quality due to a lack of common, generally agreed upon criteria for the aforesaid evaluation. Strict criteria can only be applied to narrow, specific cases, such as vocal melody in Renaissance counterpoint.…”

Transparency in Music-Generative AI: A Systematic Literature Review

“…In the music generation field, Guo et al (2023) highlight the gap of objective methods to evaluate music melody in automatic music generation, and claim the need for an objective and comprehensive method. They propose an objective melody evaluation strategy based on music theory and mathematical statics, which is able to assess the quality of a generative model and the generated music.…”