• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ³í¹®Áö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ³í¹®Áö > Çѱ¹Á¤º¸°úÇÐȸ ³í¹®Áö > Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)

Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) À½¾Ç À¯»çµµ ºñ±³¸¦ À§ÇÑ Siamese ³×Æ®¿öÅ© ±â¹Ý ±×·¡ÇÁ ÀÓº£µùÀÇ °³¼±
¿µ¹®Á¦¸ñ(English Title) Improvement of Graph Embedding Based on Siamese Network for Comparison of Music Similarity
ÀúÀÚ(Author) ¼ÛâÇå   ÀÌ¿ëÇö   ±èÇüÁÖ   Changheon Song   Yonghyun Lee   Hyungjoo Kim  
¿ø¹®¼ö·Ïó(Citation) VOL 26 NO. 11 PP. 0493 ~ 0498 (2020. 11)
Çѱ۳»¿ë
(Korean Abstract)
À½¾Ç ½ÃÀåÀÇ ¼ºÀå¿¡ µû¶ó »ç¿ëÀÚ´Â ÀϺΠÀ½¾Ç¿¡ ±¹ÇÑµÇ¾î ³ëÃâµÇ°í ¼±ÅÃÇÏ°Ô µÈ´Ù. ¸¹Àº ¼­ºñ½º´Â ¸ÞŸµ¥ÀÌÅÍ·Î ¶óÀ̺귯¸®¸¦ ±¸¼ºÇÏ¿© °Ë»ö ¹× Ãßõ ¹®Á¦¿¡ Á¢±ÙÇÏ°í ÀÖ´Ù. À̶§, »õ·Î ³ª¿À°Å³ª ÀÎÁöµµ°¡ ¾ø´Â À½¾ÇÀÇ °æ¿ì °á°ú¿¡¼­ Á¦¿ÜµÉ ¼ö ÀÖ´Ù. ÀϹÝÀûÀ¸·Î »ç¿ëµÇ´Â ¿Àµð¿À ÇÇó´Â Çػ󵵿¡ µû¸¥ Â÷¿øÀÇ º¯È­ ÆøÀÌ Å©±â ¶§¹®¿¡ CNNÀÇ ÀÔ·ÂÀ¸·Î »ç¿ëÇϱ⿡ ¾î·Á¿òÀÌ ÀÖ´Ù. º» ³í¹®¿¡¼­´Â À½¾Ç ±×·¡ÇÁ ÇÇó¸¦ ÃßÃâÇÏ°í ÀÓº£µùÇÏ¿© À¯»çµµ¸¦ ºñ±³ÇÒ ¼ö ÀÖ´Â ¸ðµ¨À» Á¦¾ÈÇÑ´Ù. ¸ðµ¨Àº ÇÇó ÃßÃâ°ú Siamese ³×Æ®¿öÅ©·Î ±¸¼ºµÈ´Ù. ÇÇó ÃßÃâ¿¡¼­´Â °¢ À½¾Ç ½ÅÈ£¸¦ ¿Àµð¿À ÇÇó·Î º¯È¯ÇÏ°í, °¢ À½¾ÇÀÇ ±×·¡ÇÁ ÇÇó¸¦ ±¸¼ºÇÑ´Ù. ÀÌÈÄ, Siamese ³×Æ®¿öÅ©¿¡¼­ °¢ ±×·¡ÇÁ ÇÇó¸¦ GCN°ú ¾îÅÙ¼Ç ±â¹ýÀ» È°¿ëÇÏ¿© ÀáÀç °ø°£À¸·Î ÀÓº£µùÇÏ°í, NTNÀ» ÅëÇØ ¼­·Î ´Ù¸¥ µÎ º¤ÅÍÀÇ À¯»çµµ¸¦ µµÃâÇÑ´Ù. ¸¶Áö¸·À¸·Î ½ÇÇèÀ» ÅëÇØ À½¾Ç ½ÅÈ£ÀÇ À¯»çµµ ºñ±³¸¦ À§ÇÑ ¿Àµð¿À ÇÇóÀÇ ±×·¡ÇÁ ÇÇó ÃßÃâÀÌ È¿°úÀûÀÎ ¹æ½ÄÀÓÀ» ÀÔÁõÇÏ¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
As the music market grows, people are exposed to and provided with selective music. Many services use metadata for building music libraries. In this situation, songs from independent labels and new artists that do not have previous information are still excluded from the result of searches and recommendations. In this paper, we focus on making the music scoring model for calculating the similarity score of two music signals. The model comprises the Siamese network and the scoring layer. The Siamese network embeds audios to small latent vectors and passes them to the scoring layer. The audio feature is difficult to use as an input to the CNN because of the dimensionality problem. Our approach is compared to previous works because it retains the sequence information of the peak frequencies in the spectrogram by transforming it into a graph. The effectiveness of the graphical approach is shown as the result of the experiment.
Å°¿öµå(Keyword) ¿Àµð¿À ÄÁÅÙÃ÷   ±×·¡ÇÁ ÀÓº£µù   ±×·¡ÇÁ Äܺ¼·ç¼Å³Î ³×Æ®¿öÅ©   Siamese ³×Æ®¿öÅ©   audio content   graph embedding   graph convolutional network   Siamese network  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå