Research on singing rendering systems design based on an active auditory perception model

这个项目的目的是调查为什么声乐即使没有歌词也有吸引力的原因。这个总目标被分解成几个子目标，包括新的研究工具开发和在国际比赛中获得最佳人工歌唱系统奖。尽管成功带来的问题比解决的问题要多，但这些目标还是实现了。首先，在2004年在日本静冈县举行的国际计算机娱乐系统会议（NIME'04）的附属活动“RENCON'04”上，人工合成人声合唱作品（摘自竹光彻的作品《小天空》）在四种合成演唱系统中获得了一等奖。该作品是使用基于直唱歌合成程序。其次，歌唱合成系统是基于本课题所发明的听觉变形算法。变形算法对语音感知和音乐感知的研究产生了重大影响，该算法目前在世界上许多研究机构中得到应用。第三，提出了一种新的“振音感”算法，使歌唱声的基本组成部分振音的变形成为可能，同时也是歌唱声变形困难的障碍。第四，根据当前研究项目所提出的“系统降级策略”，通过实际调查获得了重要经验。这些成就已在各种国际/国内会议和科学期刊上报道。这些出版物和基于STRAIGHT的新研究工具形成了以听觉和言语感知的生态学观点为特征的研究趋势。总之，这个项目取得了巨大的成功。然而，值得注意的是，即使在这个项目中取得了所有的成就，合成歌手和人类歌手之间仍然存在巨大的差距。要弥合这一差距，调查还有很大的空间。前瞻性研究项目可能需要关注从相对较少的实例中进行泛化的方法，因为根据本研究项目的经验，提供足够数量的歌声实例来充分发挥“系统降级策略”的作用通常是不切实际的。少

The goal of this project is to investigate the source of reason why vocal music is attractive even without lyrics. This general goal was broken down to several sub-goal which consists of new research tool development and winning prize as the best artificial singing system at international contests. These goals were fulfilled even though the success introduced more questions than the answered questions. Firstly, the piece of chorus with artificially manipulated synthesized voices (that is an excerpt of a composition made by Toru Takemitsu titled "small sky") won the first prize among four synthetic singing systems at RENCON'04, the satellite event of the international conference on computer based entertainment systems (NIME'04) held in Shizuoka in 2004. The piece was made using a STRAIGHT based singing synthesis program. Secondly, the singing synthesis system is based on the auditory morphing algorithm invented for this research project. The morphing algorithm made a substantial impact … More on speech perception and music perception research and the algorithm is currently used in many research institutes worldwide. Thirdly, a new algorithm called "senza vibrato" was developed to made it possible to morph vibrato that is an essential ingredient of singing voice, and at the same time, is an obstacle that made morphing of singing voice very difficult. Fourthly, important experiences were obtained by performing actual investigations based on the "systematic downgrading strategy" that was proposed to characterize the current research project. Those accomplishments were reported at various international/domestic conferences and scientific journals. Those publications and the new research tools based on STRAIGHT made a research trend that is characterized by ecological views on auditory and speech perception. In conclusion, the project was a great success. However, it is important to note that even with all the accomplishments in this project, there still remains a huge gap between synthetic singers and human singers. There is a huge room for investigations to bridge this gap. The prospective research project may need to put attentions on methods for generalization from relatively small number of instances, because, based on experiences in this research project, it is generally impractical to provide sufficient number of singing voice instances to function the "systematic downgrading strategy" in its full extent. Less