Extracting melody from audio recordings is a difficult and important research problem, because melody is one of the most important descriptors of (Western) music. Successful audio melody extraction would bridge the gap between a large number of researches made on symbolic (MIDI) data and audio, and enable a number of applications from retrieval in audio collections, segmentation, summarization, voice removal, music learning etc.
Melody extraction algorithm
Our approach to melody extraction is based on finding significant melodic fragments throughout the analyzed piece of music and clustering these fragments according to their timbral similarity. Fragments in each cluster are taken to represent fragments belonging to the same melodic line, different clusters contain different melodic lines. Holes between significant fragments within each cluster are filled-in by a shortest-path approach over all melodic fragments.
The algorithm was also submitted to MIREX 2005 Audio Melody Extraction task.
For more details, see:
Some examples (see papers for evaluation)
- Just like being in love
Source Extracted melody Source+extracted melody - Frozen
Source Extracted melody Source+extracted melody
Grouping melodic lines with similar timbre