Query by humming system for embedded platforms

Query by humming (QBH) is a method of searching for a songs in a multimedia database system that contains the melody descriptions of songs. The database of songs can be searched by hummed queries. The user hums a melody into a microphone that is connected to any handheld device and the QBH system searches the database of songs which are similar to the input query and returns the result to the user as a list of songs that match. The main objective of this paper is to propose a QBH system for embedded devices like mobile phones, MP3 players etc., To achieve the same the algorithms for the extraction of feature and their symbolic representation, from the query signal and the database are described. A novel string matching technique to compare the feature representation of query and database of songs are also described. Our retrieval database is built on 100 MIDI files and evaluated using 440 hummed samples from 40 people with different musical backgrounds. A retrieval accuracy of 83% is demonstrated for humming samples from musically untrained subjects.