Applications of Binary Classification and Adaptive Boosting to the Query-By-Humming Problem

In the “query-by-humming” problem, we attempt to retrieve a specific song from a target set based on a sung query. Recent evaluations of query-by-humming systems show that the state-of-the-art algorithm is a simple dynamic programming-based interval matching technique. Other techniques based on hidden Markov models are far more expensive computationally and do not appear to offer significant increases in performance. Here, we borrow techniques from artificial intelligence to create an algorithm able to outperform the current state-of-the-art with only a negligible increase in running time.

• Type: Conference
• Year: 2005
• Where: ISMIR
• Authors: Charles L. Parker
• Keywords: Hidden Markov Models, Information Retrieval, Interval Matching Technique, ISMIR 2005, State-of-the-art Algorithm