A Query-by-Singing System based on Dynamic Programming

This paper presents a query-by-singing system of which the comparison engine is mostly based on the concept of dynamic programming (DP). The system, known as CBMR (Content-Based Music Retrieval), facilitates the content-based song database retrieval via users' acoustic inputs. CBMR first takes the user's acoustic input from a microphone and converts it into a pitch vector. Then two levels of comparison procedures, both based on the concept of dynamic programming, are invoked to compute the similarity between the user's pitch vector and that of each song in the database. CBMR then shows a ranked result according to the computed similarity scores. We have tested CBMR extensively and found the performance is satisfactory for average people with mediocre singing capability. Our studies demonstrate the importance of using dynamic programming for elastic matching in multimedia informational retrieval.