An evaluation of music information retrieval techniques

Music or audio data is becoming an important part of Big Data spheres, especially in social platforms such as Youtube, Facebook, and Spotify. While many music information retrieval (MIR) techniques have been developed over the last decades, there has been no work on the evaluation of their performance altogether. The main reason is the lack of a common setting (i.e. no common dataset and no common metrics of success). As a result, understanding the performance implications of these techniques is challenging, since each of them has distinct characteristics. Moreover, MIR techniques have never been compared systematically, and each work often reported its superior performance generally using a limited variety of datasets or evaluation methodologies. Therefore, there is a need of common settings to test, research, and assess the advantage and disadvantage of these techniques.

In this project, we will develop a benchmark to evaluate different music information retrieval techniques. Possible applications include similarity search, recommender systems, and audio query processing.

Potential resources:

• Typke, R., Wiering, F., & Veltkamp, R. C. (2005). A Survey Of Music Information Retrieval Systems. Ismir, 153–160. Retrieved from http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.59.7958
• Wang, A. (2006). The Shazam music recognition service. Communications of the ACM, 49(8), 44. http://doi.org/10.1145/1145287.1145312
• IDI, N. (n.d.). Methods for retrieving musical information based on rythm and pitch correlations, 1–15. Retrieved from http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.128.4754&rep=rep1&type=pdf
• Yu, Y., Zimmermann, R., Wang, Y., & Oria, V. (2013). Scalable content-based music retrieval using chord progression histogram and tree-structure LSH. IEEE Transactions on Multimedia, 15(8), 1969–1981. http://doi.org/10.1109/TMM.2013.2269313
• Shen, J., Pang, H., Wang, M., & Yan, S. (2012). Modeling concept dynamics for large scale music search. In Proceedings of the 35th international ACM SIGIR conference on Research and development in information retrieval – SIGIR ’12 (p. 455). http://doi.org/10.1145/2348283.2348346

Deliverables: datasets, baselines, evaluation methodologies, trained models, accuracy results, web/mobile app (OPTIONAL)

The project requires programming skills in Python, C++, Matlab, etc. Competent knowledge and experience in information retrieval, query processing, high-dimensional data. Pro-active in learning and trying new things, everyday.