Abstract:
Since the development in digital technology, music has a lead role in the leading tech-
nological evolution. Music can be thought as basic human need. As huge amount of
data is required to store and transfer audio signal, challenges arise as how e ciently
one can store audio signal without almost no changes in its quality. With the use of
cloud backup one can have own music collection, but this leads to wastage of band-
width as the same music content can repeat for persons. So, to index and search these
vast amounts of data becomes a challenging task. In melody extraction, the melody
( pre-dominant fundamental frequency of a music signal) is extracted from the music
which then gives salient applications like indexing and searching music content, query
by humming (QBH), voicing detection, voice separation, genre classi cation (like Rock,
Pop, Indian etc...), music teaching (by extracting di erent beat frequencies), games, se-
curity systems, karaoke system and more. So many melody extraction algorithms have
been developed, but most of the algorithms depend on a type of music where for speci c
music it can work e ectively. Melody extraction can be vocal or instrumental. Vocal
melody extraction algorithms extract the human voice (Voicing detection) whilst instru-
mental includes detection of particular instrument sound like tabla, guitar,
ute etc.
As synchrosqueezed wave packet transform gives very accurate time-frequency represen-
tation, the need for extra steps to calculate melody is eliminated and directly giving
nal melody. Proposed melody algorithm shows an improved accuracy compared some
existing methods.
