Its highly unlikely that we iterate over motif lengths, sothe idea would be to find a sensible length and then extend the motif once we find suitable repetition.

• which algorithm should be used for motif length extension
• what should be constraints on length extension?

Information about itself

• time stamp of the motif
• file name of the motif

Other information

• Motif pair information, ID of the motif pair
• Distance with the motif pair
• Information about the dataset, what was the dataset and what downsampling and what preprocssing etc

• How to select motifs within a song, threshold or number of motifs
• How do we find threshold while searching for occurances of discovered seed motifs
• Do we see any relation with the motif characteristics?
• Should we start atleast with a sensible threshold to make discovery faster?

• decide length of motif
• decide sampling rate of pitch
• decide criterion for removing flat regions
• decide criterion for removing solo percussion regions
• decide hop size while generating subsequences
• do not cut across silence region while generating subsequences
• fast way to separate breath phrases

• Should DTW be used with some local constraints or not?
• Should we use DTW subsequence version or point to point version. If p2p then how do we handle tempo variations. Should we have multiple subsequences with linear stretching?
• Should we use weights for insertion and deletion?
• Should we path normalize distances? In which case lowerbounds might not work
• Should we have early abandoning enabled or disabled