Project Ideas

Possible Topics

Details

Identify beginning and end of phrase
Definitely easiest with monophonic music
for polyphonic music, phrase identification could follow melody transcription
probably need more features than what the melody transcriber gives. Mix original audio with "cooked" output from melody transcriber

Naive approach: look for loudest spectral peak
Look for harmonically-related peaks to find a fundamental pitch
This seems like a relatively obvious/well researched problem. What about related issues?
- catching presence/absence of main melodic line. what do melody-extractors do when the melody takes a break?
- identify changes in instrumentation
References
- Durey, A. S., and Clements, M. A. Features for Melody Spotting Using Hidden Markov Models. In Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, 2002.
- Herrera, P., Amatriain, X., Batlle, E., and Serra. X. Towards Instrument Segmentation for Music Content Description: A Critical Review of Instrument Classification Techniques. In Proceedings of the International Symposium on Music Information Retrieval, 2000
- http://www.music-ir.org/mirex/2005/index.php/Audio_Melody_Extraction

Perhaps focusing on following an inexact performance
performance mistakes or semi-improvised parts?
polyphonic audio sources seem less explored, but in a performance environment it seems that you could easily put a microphone on a single instrument to get a monophonic signal for tracking, so how necessary is it?
What kind of training data would I need? Not much available for a specific live performance