Computer science problems related to music?

Question

Are there any CS problems, preferably open, that are related to music or musical theory somehow? I would think of problem with musical notation but also probabilities when randomizing according to a scale or a tonality or general what is considered harmony in frequencies and physics, electromagnetism and waveforms.

Can you give examples of the area I want to know of?

For instance, given an algorithm that guesses a melody, how successful will the melody be in resembling an artist or likewise decision problem that could be feasible or what do you think?

Answer 1

It is easy to think of many different problems:

1. Converting sound to scores.
2. Converting scores to sound.
3. Automatic music composition.
4. Music analysis, for example key recognition.
5. Music classification, such as style recognition and artist recognition.

All of these things are actually being done.

Answer 2

music is a nice/neat/natural way to approach some very advanced CS algorithms/applications etc. and there has been a large burst of innovation and crosspollination between the two areas in the last decade especially in computer music software.

one important area involves attempting to separate different channels from a mixed sound. in other words suppose separate tracks are mixed together into a single track (separate instruments, vocals, etc). its actually a very advanced statistical/AI problem of separating out the original tracks from the mixed track. another basic application is attempting to remove vocals only from music; there are devices advertised to do this, some of them highly engineered and expensive, called vocal removers. this same problem arises in other guises in statistics. my understanding is that ICA, independent component analysis can be used to solve this problem.

another interesting area of application of CS to music related to vocals, enjoying a massive surge in use in the last decade: pitch correctors or so-called autotune which adjust vocal pitch to discrete, on-key increments. Autotune has an origination in scientific/statistical/mathematical analysis; from wikipedia:

Auto-Tune was initially created by Andy Hildebrand, an engineer working for Exxon. Hildebrand developed methods for interpreting seismic data and subsequently realized that the technology could be used to detect, analyze, and modify the pitch in audio files.[3]

also, audio plugins for effects, or "effects boxes", often involve many highly advanced CS/signal processing concepts, closely connected to algorithmic and mathematical theory, such as Fourier Transforms, etcetera. another common effect with some algorithmic theory/complexity is reverb which has roots in mathematical convolution.

ref [3] from wikipedia with more on autotune & Hildebrand: The Gerbils Revenge / Auto-Tune corrects a singer’s pitch. It also distorts—a grand tradition in pop., Sasha Frere-Jones, New Yorker 6/9/2008

Answer 3

There is a machine information retrieval (MIR) field that is concerned with retrieving information from music. You may find interesting paper "Machine Learning Approaches for Music Information Retrieval" by Tao Li et al. which describes several problems in MIR and corresponding machine learning approaches:

• Emotion detection in music.
• Music style identification.
• Music recommendation (for example, for playlist generation).

Another good paper is "Contextual music information retrieval and recommendation: State of the art and challenges" by Marius Kaminskas et al. which also describes the already mentioned MIR problems and:

• Querying music by example
• Querying music by humming

PhD thesis "Music Recommendation and Discovery. In the Long Tail" by Òscar Celma Herrada is a terrific resource and I personally enjoyed the slides.

Finally, MIR summer school will be soon organized. The registration is already closed, but you may find interesting the materials of this school.