General information

Master in Sound and Music Computing, DTIC, UPF
2010/2011. Second term.
Instructor: Emilia Gómez  (emilia.gomez at upf dot edu)
Credits: 5 ECTS
Schedule: Wednesday 16:30-18:00 (lecture, room 52109) and Friday 15:00-16:30 (hands-on exercices, room 54009)

Course presentation

This course focuses on methodologies and techniques for the automatic characterization of musical audio in terms of different facets (e.g. melody, harmony, rhythm, timbre, and spatial location), temporal scopes and abstraction levels (from low-level to semantic descriptions). Special emphasis is given to state-of-the-art signal processing methods for audio content description, interdisciplinary research and music information retrieval.


To take this course it is desirable to have an engineering background, to have taken some courses in Mathematics at the undergraduate level, like Algebra and Calculus, and also to be familiar with basic signal processing concepts. Programming experience and/or formal music knowledge is also desirable.

Competences to be acquired

General: Mathematic and analytic skills; Ability to find and use information to solve a given problem; Ability to communicate in English; Ability to work in a team; Ability to work in an interdisciplinary field;

Specific: Develop methods for describing audio signals; Understand current state of the art analysis techniques; Apply signal processing methodologies to solve practical problems related to audio and music analysis; Combine musical knowledge and audio signal processing to perform musical significant descriptions.


  1. Levels and facets of music content.
  2. Temporal vs frequency representation of audio signals.
  3. Timbre and instrument description and classification.
  4. Pitch estimation and melodic description.
  5. Harmonic and tonal description.
  6. Computational methods for rhythmic analysis of audio.
  7. Music structural analysis.
  8. Bridging the semantic gap.
  9. Music classification and comparative analysis using machine learning techniques.
  10. Application contexts for music content analysis in Music Information Retrieval (MIR) 


Class format and evaluation method

The course takes place in the 2nd term of the year (10 weeks), and it is organized in two types of session each week:

The first session mainly consists on lectures, although it will sometimes include seminars or presentations by students. The second one focuses on practical work with computers.

Homework: each week, all students are expected to review the lecture material and to work on a set of questions proposed at the lectures and small programming assignments proposed at the hands-on sessions.

The labs should be carried out in teams of no more than two students. They will consist on implementing a set of descriptors related to the different musical facets. They will be implemented in Matlab and will sometimes use existing developed programs. Students will present their results during the lab sessions.

Final exam: the student will be asked to discuss some of the topics of the course at a final exam.

Evaluation: the evaluation of the course is based on the following items:








12th-14thJanuary 2011

Course introduction (Initial evaluation).

Content-based sound and music description

Low-level descriptors.

Presentation of Lab1: Low-level features and timbre.


19th-21st January 2011

Timbre description and instrument classification.




26th-28th January 2011


NO CLASS (To be scheduled later)

Work on Lab1



February 2011

Pitch and melodic description.

Presentation of results for Lab1



February 2011

Harmony and tonality

Presentation of Lab2:Melody.

Lab2 additional material: Yin: "private" folder for yin including mac mex files; Singing voice melodies.; Onset detection code.




February 2011





February 2011

Presentation of results for Lab2


Lab3: Chord and key. Presentation.lab description.

Lab 4: Tempo induction. Presentation.lab description.



March 2011

Similarity computation for structural analysis and retrieval.

Lab3, Lab4





March 2011

Comparative analysis and automatic classification (Mood classification).

Comparative analysis and automatic classification (Genre classification).

Presentation of results for Lab3



March 2011

Content-based transformations.


Presentation of results for Lab4

Course evaluation


Course material

A moodle page is available for the students, including updated schedules, course slides and lab material.  

Lab format: lab reports should be between 4 and 6 pages long, in PDF format,  and conform to the guidelines in the following ISMIR word and LaTeX templates.

Core technical papers from current literature.

Additional material

Complementary References

Relevant Software