We have constructed a dataset consisting of 100 manually created microtonal music tracks. It is intended for use in research purposes.
[Project page]

This is a VST plugin utilizing deep generative models. This plugin suggests continuations of melodies as users input them on the piano roll. The suggestions are based on a deep learning model constructed with the user's input data as training data.

In this study, we propose a method for the automatic design of quantum circuits for melody generation using quantum computers. We successfully created quantum circuits that realize note transitions of the melody serving as teacher data by utilizing a genetic algorithm.

In this study, we have expanded the capabilities of the piano roll, which until now could only accept input of notes in the 12-tone equal temperament. It now allows for the input of melodies using notes from any temperament, including microtones. Accompanying this, we have implemented an interface that enables users to design their own original temperaments. We propose features such as the automatic generation of temperaments using frequencies contained in specific sounds, such as bird songs.

In this study, we propose a technique for distributed representation (vectorization) to handle musical melodies numerically on a computer. In order to apply Word2vec, a distributed representation technique for words, to musical melodies, we suggest a method for dividing musical melodies into words based on a rule called GPT, and a modeling technique using this method.

In this project, we focus on the typical DJ technique of smoothly transitioning between tracks, known as "mixing." We propose an automation system for this technique and a support system for DJ performances utilizing it.

In this project, we focus on the redundancy and continuity of video audio and visuals. We propose a method to expand or shrink the length of a video by selectively thinning out and inserting frames. With this approach, viewing time, which traditionally had to be adjusted to the length of the video, can now be adapted to the viewer's available time, making it possible to enjoy the video in a more flexible manner.

In this research, we propose a technique for transitioning as seamlessly as possible between two video scenes shot in the same location. By using a hierarchical graph cut, we have achieved a method to transition scenes at the pixel level for each frame of the video.

In this project, we propose new ways of viewing and generating content by blending the worlds within video content with the real world. For example, this allows for possibilities such as dancing with characters in a video or stepping into a film to practice acting.

In this project, we have developed a system to create a composite song by blending multiple tracks together. This system aims to enable a society of user-generated content where even general listeners without songwriting experience can create new songs suited to their preferences, in continuation with their music-listening activities.

In this study, we proposed a facial recognition technique that leverages the characteristics of video content and achieved the extraction of the same individual from multiple videos. By combining singing scene estimation technology (acoustic signal processing), we proposed a method for extracting singing individuals within video content. Based on facial images labeled with names, we are automatically annotating video content with people's names.

This system generates new music videos by automatically cutting and pasting footage from a video database to match an inputted music. This system aims to assist users with no video editing experience in easily creating new music video content. By doing so, it helps push forward the first step towards making everyone a creator.