The GenMusic project aims at building machine learning models for automatic music generation. We are in particular interested in generating music in the style of pop/rock songs. Our research covers the following topics: melody generation, melody harmonization, lead sheet generation and arrangement, drum pattern generation, and multitrack music generation.
Understanding the surrounding environment and ongoing events through acoustic cues is important in applications such as smart surveillance, smart city, smart car, and smart factory. This project is about building such machine learning models for sound event recognition. In addition to accuracy, our model is noise-robustness, and can be deployed on edge devices.
Given an audio mixture composing of the sounds from various sources/instruments, we build a source separation model to isolate the sounds from each individual source. For example, to separate the singing voice from the instrumental background. Here is a demo website:
We are building models to manipulate musical audio like DJs, e.g. by combing different music pieces. We have built the following models thus far:
We have been working on various topics of recommender systems, including collaborative filtering based recommendation, content-based recommendation, and context-aware recommendation.