Music Content Analysis by Machine Learning

---

This page provides the materials for a short course on Music Content Analysis. This course was first presented at the Music Technology Group of the Universitat Pompeu Fabra in Barcelona from March 18-21, 2003.

Much of this material is adapted from my course at Columbia on Speech and Audio Processing and Recognition, but with the addition of many music-specific examples.

Slide Packs

The course is presented in three, 3-hour sessions. Here are the slide packs for each session (in PDF format).

• Session 1: Basics of pattern classification: MLP neural networks, Bayesian classification, Gaussian models. Application to singing detection in music.
• Session 2: Sequence recognition: Brief overview of current speech recognition technology (where these ideas originated); introduction to hidden Markov models. Application to chord sequence recognition in music
• Session 3: Musical applications: Music transcription, summarization, artist/genre classification, and similarity/recommendation browsing.

A condensed version of this material was presented as a single 75 minute lecture as part of the Johns Hopkins CLSP Workshop 2003. The slides are here: Pattern Recognition Applied to Music Signals. These slides serve as an introduction for the self-paced practical below.

Matlab resources

The course includes illustrations mainly based in Matlab. Below are some Matlab diaries and other resources related to the material:

• Diary of examples on basic pattern recognition. The diary and the associated m-scripts are available in the Matlab directory
• Diary of Singing classification using MLPs and GMMs. This is based on the excellent and free Netlab machine learning toolbox for Matlab.

For the July 2003 session at Johns Hopkins CLSP, I structured this material into a 3-hour self-paced Matlab practical in which you go through building and evaluating singing detection systems using GMMs and Neural Nets:

• Music Content Analysis: A Practical Investigation of Singing Detection

Acknowledgment

This material is based in part upon work supported by the National Science Foundation under Grant No. IIS-0238301. Any opinions, findings and conclusions or recomendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation (NSF).