Participants 2022
CCOM-515
AI Song Contest 2022 / participantsTEAM / CCOM-515
SONG / AI Maria
TEAM MEMBERS / Sander Wood, Nancy Zhang
About the TEAM
All team members are from the Department of Music AI and Information Technology, Central Conservatory of Music.
Sander is a PhD student who majors in music artificial intelligence with Prof. Sun Maosong and Prof. Yu Feng. His main research interest is the application of deep learning algorithms in the symbolic music domain. He studied at Xinghai Conservatory of Music from 2015 to 2019 as a bachelor, majoring in piano performance. In 2019, he was admitted to the School of Computer Science and Engineering at Sun Yat-sen University for his master's degree, majoring in software engineering.
Nancy is a PhD student in electronic music composition, class of 2020, under the supervision of Prof. Xiaobing Li. Her main research interest is the application of artificial intelligence technology in the practice of electronic music composition.
About the SONG
AI Maria is a symphony which is co-composed with DeepChoir based on Ave Maria. DeepChoir is a melody choralization system which trained and validated on JSB Chorales Dataset. The melody of Ave Maria is composed by Charles Gounod, and its piano accompaniment is from Prelude No. 1 in C major, BWV 846. This music is a soulful homage to classical music, filled with Baroque elements, which is conceived as a genuine collaboration between humans and machines spanning centuries.
The code and other generated samples of DeepChoir, and the dataset we use are available in https://github.com/sander-wood/deepchoir. The paper will be updated around mid-2022.
About the HUMAN-AI PROCESS
Melody choralization has long been closely associated with JSB Chorales. Previous neural network-based systems rarely focused on chorale generation conditioned on a chord progression, and none of them realised controllable melody choralization. To enable neural networks to learn the general principles of choral composition, we first designed a music representation that encoded chord symbols for chord conditioning. We then propose DeepChoir, a melody choralization system, that can generate a four-part chorale for a given melody conditioned on a chord progression. Furthermore, we applied gamma sampling in the melody choralization task, and users can control the extent of harmonicity for the chorale generated by DeepChoir.
Ave Maria is a well-known and well-recorded setting of a Latin prayer. The piece includes a melody by the French Romantic composer Charles Gounod that he superimposed over Prelude No. 1 in C major, BWV 846, which is composed by Bach. We aimed to further rearrange it into a symphony with the assistance of DeepChoir.
This symphony is created as follows. We first chordify the piano accompaniment part to get its chord progression. DeepChoir (a melody choralization system) then generates a four-part SATB choir based on the extracted chord progression and the melody. During this process, we can adjust the harmonicity to guide the system in their intended musical direction. Finally, we rearrange the fine-tuned generated choir into a symphony, which be named as “AI Maria”.