Extraction of predominant melody from the musicalperformances containing various instruments is one of the mostchallenging task in the field of music information retrieval andcomputational musicology. This paper presents a novel frameworkwhich estimates predominant vocal melody in real-timeby tracking various sources with the help of harmonic clusters(combs) and then determining the predominant vocal source byusing the harmonic strength of the source. The novel on-lineharmonic comb tracking approach complies with both structuralas well as temporal constraints simultaneously. It relies uponthe strong higher harmonics for robustness against distortionof the first harmonic due to low frequency accompaniments, incontrast to the existing methods which track the pitch values. Thepredominant vocal source identification depends upon the novelidea of source dependant filtering of recognition score, whichallows the algorithm to be implemented on-line. The proposedmethod, although on-line, is shown to significantly outperformour implementation of a state-of-the-art offline method for vocalmelody extraction. Evaluations also show the reduction in octaveerror and the effectiveness of novel score filtering technique inenhancing the performance.
|Journal||IEEE Transactions on Audio, Speech and Language Processing|
|Publication status||Published - Mar 2013|
Arora, V., & Behera, L. (2013). On-Line Melody Extraction From Polyphonic Audio Using Harmonic Cluster Tracking. IEEE Transactions on Audio, Speech and Language Processing, 21(3), 520-530. https://doi.org/10.1109/TASL.2012.2227731