A Novel HMM Approach to Melody Spotting in Raw Audio Recordings

This paper presents a melody spotting system based on Variable Duration Hidden Markov Models (VDHMM’s), capable of locating monophonic melodies in a database of raw audio recordings. The audio recordings may either contain a single instrument performing in solo mode, or an ensemble of instruments where one of the instruments has a leading role. The melody to be spotted is presented to the system as a sequence of note durations and music intervals. In the sequel, this sequence is treated as a pattern prototype and based on it, a VDHMM is constructed. The probabilities of the associated VDHMM are determined according to a set of rules that account (a) for the allowable note duration flexibility and (b) with possible structural deviations from the prototype pattern. In addition, for each raw audio recording in the database, a sequence of note durations and music intervals is extracted by means of a multi pitch tracking algorithm. These sequences are subsequently fed as input to the co...

• Type: Conference
• Year: 2005
• Where: ISMIR
• Authors: Aggelos Pikrakis, Sergios Theodoridis
• Keywords: Audio Recording, Information Retrieval, ISMIR 2005, Note Durations, Raw Audio Recording