Lecture: Audio Decomposition¶
After working through the material of this lecture, you should be able to answer the following questions:
- What is the general goal of source separation?
- What is the aim of harmonic–percussive separation? Which signal properties are exploited?
- What is the goal of melody separation?
- What is main idea of nonnegative matrix factorization? How can this technique be applied for audio decomposition?
- What is the musical interpretation of template and activation matrices?
Reading Assignments¶
- Chapter 8, Müller, FMP, Springer 2021
- Introduction to Chapter 8
- Section 8.1: Harmonic–Percussive Separation
- Introduction to Section 8.1
- Section 8.2: Melody Extraction
- Introduction to Section 8.2
- Section 8.3: NMF-Based Audio Decomposition
- Introduction to Section 8.3
Slides¶
- Audio Decomposition: Slides (PDF), Handouts (6 slides per page) (PDF)
Videos¶
Here is a selection of videos related to tempo and beat.
- Demo: Score-Informed Audio Decomposition and Applications (3:51)
Piano music; note-wise decomposition; user interface; audio editing; pitch shifting; residual signal; left and right hand decomposition - Source Separation (53:40)
Lecture by Dan Ellis; sources; mixture; perception; spatial filtering; time–frequency masking; model-based separation - Spleeter from Deezer as a source separation tool for music tracks (4:45)
Machine learning framework for source separation of music; Michael Jackson: Billie Jean (0:01); Modern Talking: Brother Louie (1:03); Katy Perry: Dark Horse (2:22); Grimes: REALiTi(3:02); Scatman John: Scatman's World (3:54)
Question & Answer Session¶
