In this paper, we present a novel approach for music identification task aimed at proving the ability to identify a song by recorded song snippets. By combining Y. Ke’s feature extracting method [1, 2] with PostgreSQL user-defined functions [3, 4, 5]], our system proves as an effective search strategy for the field. We construct training data sets in a noisy environment and compare the search speed and the search accuracy of the system with Y. Ke’s system. Experiment results show that our system is more powerful with the accurate retrieval ability of 98% on a database of 600 songs and the search speed is 3.6 times faster than Y. Ke’s system.
Readership Map
Content Distribution
In this paper, we present a novel approach for music identification task aimed at proving the ability to identify a song by recorded song snippets. By combining Y. Ke’s feature extracting method [1, 2] with PostgreSQL user-defined functions [3, 4, 5]], our system proves as an effective search strategy for the field. We construct training data sets in a noisy environment and compare the search speed and the search accuracy of the system with Y. Ke’s system. Experiment results show that our system is more powerful with the accurate retrieval ability of 98% on a database of 600 songs and the search speed is 3.6 times faster than Y. Ke’s system.
