Source separation refers to the process of extracting constituent source signals from one or more mixture signals. Access to high-quality musical sources separated from an audio mixture may facilitate downstream audio processing and music information retrieval tasks, such as audio enhancement and equalization, automatic music transcription, and content-based audio retrieval. In this project, we consider four-part chorales performed by predominantly-harmonic musical instruments, such as the bassoon and clarinet. The tones produced by such instruments are often well-approximated by a superposition of sinusoids oscillating at a fundamental frequency and its harmonics. We exploit musical structure to track the time-varying fundamental frequency of each source. Using these estimated fundamental frequency trajectories, we design time-varying filter banks to extract each source.

I’m looking forward to tackling an open-ended signal processing problem that I find fascinating. I hope to get more experience not only with conducting independent research, but also with communicating technical concepts clearly and concisely.