Music and Audio Computing Lab

Main menu: Home | People | Research | Publications |

Current Projects

Content-based Music Retrieval

In the recent past, music has become ubiquitous as digital data. The scale of music content in online music and video sharing services has significantly increased and we can readily access to them. This posed challenges in terms of efficient and effective content organization, search and recommendation. In this project, we explore machine learning algorithms, particularly unsupervised feature learning and deep learning, to automatically annotate music in terms of genre, mood, instruments, artist and other music descriptions.

Singing Voice Analysis in Polyphonic Music

In popular music, singing voice is the central sound source that determines the song quality, as it conveys melody, lyrics, emotion and humanity with its high expressivity. In this project, we investigate algorithms to analyze various characteristics of vocal sounds from low-level acoustic features (e.g. pitch and timbre) to high-level word descriptions of voice. In particular, we focus on extracting the information in polyphonic settings.

Intelligent System for Classical Music Listening

Classical music often contains complicate musical structure and text compared to other genres of music. In this topic, we search for technical methods that can help listeners to understand and enjoy the classical music. One of the representative examples is web-based listening interface that provides a score of the piece with its various recordings, which are synchronized with the score. We also work on analyzing performer’s interpretation of the music, so that we can explain the differences among performances in a quantitative way.

Phrase-level Synthesis of Korean Traditional Musical Instruments

Sample-based synthesis is the most popular method to generate realistic musical sounds in modern synthesizers. These days many of commercial software synthesizers provide a large number of samples in the form of musical phrases beyond individual notes for more convenient music making. In this project, we develope algorithms to process or generate samples in the phrase level, targeting for Korean traditional musical instruments.

Music Similarity and Plagiarism

Determining music plagiarism and music copyright infringement has always been controversial; there are no absolutely quantified metrics that measure the similarity between two songs. In this topic, we explore the relationship between computational similarity and perception of similarity. By comparing similarities from different measures using audio, symbolic and human data, we investigate how several musical elements affect musical plagiarism judgment.

Expressive Musical Interaction on Multi-Touch Devices

Multi-touch devices such as smartphones and tablet PCs provide a wide experimental space for exploring musical interaction. While most musical instruments implemented on these devices focus on mimicking their real-world counterpart such as the piano and the guitar, we believe that alternate gestures exclusively designed for multi-touch devices can further expand the accuracy and expressiveness of touchscreen music.

Past Projects

Audio Feedback Systems for Evolutionary Sound Generation

This project presents audio-feedback systems with high-level adaptive control toward chosen sonic features. Users control the system by selecting and changing feature objectives in real-time. The system has a second-order structure in which the internal signal processing algorithms are developed according to an evolutionary process. A prototype is evaluated experimentally to measure changes of audio feedback depending on the chosen target conditions.