Professor KAIST Daejeon, Taejon-jikhalsi, Republic of Korea
Abstract: Sample-based music creation has become a mainstream practice. One of the key tasks in the creative process is searching desired samples in the large collections. However, most commercial packages describe the samples using metadata, which is limited to explain subtle nuances in timbre and style. Inspired by music producers who often find instrument samples with a reference song, we propose a query-by-example scheme that takes mixed audio as a query and retrieves single audio samples. Our method is based on deep metric learning where a neural network is trained to locate single audio and their mixtures closely in the embedding space. We show that our model successfully retrieves single audio samples given mixed audio query in various evaluation scenarios.