文章摘要

文章信息/Info

Title:: Music synthesis method of tinnitus rehabilitation based on BPE and automatic updating Markov chain

Author(s):: FANG Yiming; HE Peiyu; FANG Ancheng; DENG Mingyue; PAN Fan; （College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China）

Keywords:: Byte pairen encoding; Automatic updating Markov chain; Music block; Generating of music; Tinnitus treatment; Self-similarity

摘要:: 为弥补现有耳鸣治疗方法的模型复杂，且合成音乐具有较强拼接痕迹的问题，本研究基于字节对编码与自更新马尔科夫链提出一种新的音乐合成方法用于耳鸣治疗。首先，使用字节对编码将原始音乐主旋律声部的音符、和弦按照顺序分割成音乐块；最后将音乐块作为状态，利用自更新马尔科夫链生成无限长、不重复且与原始曲风相似的音乐。结果表明，该方法合成的音乐保留了原始音乐的特点，听感自然、无拼接痕迹且不完全重复。该方法对耳鸣治疗具有一定的参考价值。

Abstract:: In view of the complexity of existing tinnitus treatment methods and the strong stitching trace of synthetic music,we proposed a new music synthesis method based on byte pair encoding and self-updating Markov chain for tinnitus treatment. Firstly, the notes and chords of the main melody of the original music were divided into music blocks by byte pair encoding. Finally, the music block was regarded as the state, and the self-updating Markov chain was used to generate music of infinite length, non-repetition and similar to the original music style. The results showed that the synthesized music not only retained the characteristics of the original music, but also had the advantages of natural hearing, no trace of splicing and incomplete repetition. This method has certain reference value for the treatment of tinnitus.