Song Similarity Classication
2013 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE credits
Student thesis
Abstract [en]
The purpose of this study was to investigate the possibility of automatically
classifying the similarity of song pairs. The machine learning algorithm
K-Nearest Neighbours
, combined with both bootstrap aggregating and
an
attribute selection classier, was rst trained by combining the acoustic
features of 45 song pairs extracted from the Million Song Dataset with usersubmitted
similarity for each pair. The trained algorithm was then utilized
to predict the similarity between 50 hand-picked and about 4000 randomly
chosen pop and rock songs from the Million Song Dataset.
Finally, the algorithm was subjectively evaluated by asking users to identify
which out of two randomly ordered songs, one with a low and one with
a high predicted similarity, they found most similar to a target song. The
users picked the same song as the algorithm 365 out of 514 times, giving the
algorithm an accuracy of 71%.
The results indicates that automatic and accurate classication of song
similarity may be possible and thus may be used in music applications. Further
research on improving the current algorithm or nding alternative algorithms
is warranted to draw further conclusions about the viability of using
automatically classied song similarity in real-world applications.
Abstract [sv]
Syftet med denna studie var att undersoka huruvida det ar mojligt att automatiskt
rakna ut hur lika tva latar ar. I studien anvandes maskininlarningsalgoritmen
k narmaste grannar
tillsammans med bootstrap aggregering och
en klassicerare som sallar bort ovidkommande egenskaper. Algoritmen
tranades forst genom att kombinera ett ertal akustiska parametrar med
anvandares likhetsbedomningar for 45 latpar skapades genom att kombinera
10 latar uttagna fran The Million Song Dataset med varandra. Den tranade
algoritmen anvandes sedan for att rakna ut likheten mellan 50 handplockade
och ungefar 4000 slumpmassigt valda pop- och rocklatar fran the Million
Song Dataset.
Avslutningsvis utvarderades resultaten genom en andra fas av anvandartestning.
Anvandare blev ombedda att lyssna pa en mallat, en av de 50
handplockade latarna, foljt av en av de latar som algoritmen matchat som
mycket lik och en lat som den matchat som mycket olik, i slumpmassig
ordning. Anvandaren ck sedan valja vilken av de tva latarna som tycktes
likna mallaten. Algoritmen och anvandaren valde samma lat i 365 av 514
fall, vilket ger algoritmen en trasakerhet pa 71%.
Resultaten tyder pa att det kan vara mojligt att utveckla en algoritm
som automatiskt kan klassicera likhet mellan latar med hog precision och
darmed skulle kunna anvandas i musikapplikationer. Ytterligare utveckling
av algoritmen, eller forskning pa alternativa algoritmer, ar nodvandigt for
att kunna dra vidare slutsatser om hur anvandbart automatisk uppskattning
av latlikhet ar for verkliga tillampningar.
Place, publisher, year, edition, pages
2013.
Series
Kandidatexjobb CSC ; K13037
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:kth:diva-134955OAI: oai:DiVA.org:kth-134955DiVA, id: diva2:668893
Educational program
Master of Science in Engineering - Computer Science and Technology
Supervisors
Examiners
2013-12-132013-12-022022-06-23Bibliographically approved