Unified Algorithm for Melodic Music Similarity and Retrieval in Query by Humming
Query by humming (QBH) is an active research area since a decade with limited commercial success. Challenges include partial imperfect queries from users, query representation and matching, fast, and accurate generation of results. Our work focus is on query presentation and matching algorithms to reduce the effective computational time and improve accuracy. We have proposed a unified algorithm for measuring melodic music similarity in QBH. It involves two different approaches for similarity measurement. They are novel mode normalized frequency algorithm using edit distance and n-gram precomputed inverted index method. This proposed algorithm is based on the study of melody representation in the form of note string and user query variations. Queries from four non-singers with no formal training of singing are used for initial testing. The preliminary results with 60 queries for 50 songs database are encouraging for the further research.
KeywordsQBH Music similarity Pattern matching Information retrieval
The authors gratefully acknowledge the support by MKSSS’s Cummins College of Engineering for providing experimental setup and the efforts by our UG students Aditi Pawle, Snehal Jain, Sonal Gawande, and Sonal Avhad for the active help in preparing query samples and experiments. Volunteer singer’s contribution for generating queries is highly appreciable. We would like to thank Dr. Sahasrabuddhe H. V. for his valuable suggestions and inputs related to musical knowledge and experiments.
- 1.Kumar P, Joshi M, Hariharan S, Rao P: Sung note segmentation for a query-by-humming system. Intl Joint Conferences on Artificial Intelligence IJCAI (2007).Google Scholar
- 2.Salamon J, Serra J, Gómez E: Tonal representations for music retrieval: from version identification to query-by-humming. International Journal of Multimedia Information Retrieval. 2(1) pp 45–58, (2013).Google Scholar
- 3.Ruan L, Wang L, Xiao L, Zhu M, Wu Y: A Query-by-Humming System based on Marsyas Framework and GPU Acceleration Algorithms. Appl. Math. pp 261–72, Feb (2013).Google Scholar
- 4.Chandrasekhar V, Sharifi M, Ross DA: Survey and Evaluation of Audio Fingerprinting Schemes for Mobile Query-by-Example Applications. ISMIR Vol. 20, pp. 801–806 (2011).Google Scholar
- 5.Molina E, Tardón LJ, Barbancho I, Barbancho AM: The Importance of F0 Tracking in Query-by-singing-humming. In ISMIR pp. 277–282, Nov (2014).Google Scholar
- 6.Gulati S, Serra J, Serra X.: An evaluation of methodologies for melodic similarity in audio recordings of indian art music. In Acoustics, Speech and Signal Processing (ICASSP), IEEE International Conference pp. 678–682 Apr, (2015).Google Scholar
- 7.Liu NH: Effective Results Ranking for Mobile Query by Singing/Humming Using a Hybrid Recommendation Mechanism. IEEE Transactions on Multimedia. pp 1407–20, (2014).Google Scholar
- 8.Wang CC, Jang JS: Improving query-by-singing/humming by combining melody and lyric information. IEEE/ACM Transactions on Audio, Speech, and Language Processing. pp 798–806 (2015).Google Scholar
- 9.Liu, N. H: Effective Results Ranking for Mobile Query by Singing/Humming Using a Hybrid Recommendation Mechanism. IEEE Transactions on Multimedia, 1407–1420(2014).Google Scholar
- 10.Ramesh V: Exploring Data Analysis in music using tool praat. ICETET. IEEE International Conference, pp. 508–509, (2008).Google Scholar
- 11.Makarand, Velankar, and H. V. Sahasrabuddhe: Novel Approach for Music Search Using Music Contents and Human Perception, IEEE International Conference on Electronic Systems, Signal Processing and Computing Technologies ICESC, (2014).Google Scholar