Sign In

Explainable AI in Speaker Recognition -- Making Latent Representations Understandable

์ž‘์„ฑ์ž
  • Haebom
์นดํ…Œ๊ณ ๋ฆฌ
Empty

์ €์ž

Yanze Xu, Wenwu Wang, Mark D. Plumbley

๐Ÿ’ก ๊ฐœ์š”

๋ณธ ๋…ผ๋ฌธ์€ ์Šคํ”ผ์ปค ์ธ์‹ ์‹ ๊ฒฝ๋ง์ด ํ•™์Šตํ•˜๋Š” ์ž ์žฌ ํ‘œํ˜„ ๊ณต๊ฐ„ ๋‚ด์—์„œ ๊ณ„์ธต์  ํด๋Ÿฌ์Šคํ„ฐ๋ง ํ˜„์ƒ์„ ๋ถ„์„ํ•˜๋Š” ์ƒˆ๋กœ์šด XAI ๋ฐฉ๋ฒ•๋ก ์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ์ œ์•ˆ๋œ Hierarchical Cluster-Class Matching (HCCM) ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ SLINK ๋ฐ HDBSCAN์œผ๋กœ ์ƒ์„ฑ๋œ ๊ณ„์ธต์  ํด๋Ÿฌ์Šคํ„ฐ๋ฅผ ์˜๋ฏธ๋ก ์  ํด๋ž˜์Šค์™€ ๋งค์นญํ•˜์—ฌ ํ•ด์„ ๊ฐ€๋Šฅ์„ฑ์„ ๋†’์ž…๋‹ˆ๋‹ค. ๋˜ํ•œ, Liebig score๋ฅผ ํ†ตํ•ด ํด๋Ÿฌ์Šคํ„ฐ์™€ ์˜๋ฏธ๋ก ์  ํด๋ž˜์Šค ๊ฐ„์˜ ๋งค์นญ ์ •๋„๋ฅผ ์ •๋Ÿ‰ํ™”ํ•˜๊ณ  ์ œ์•ฝ ์š”์ธ์„ ํŒŒ์•…ํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
์Šคํ”ผ์ปค ์ธ์‹ ๋ชจ๋ธ์˜ ์ž ์žฌ ํ‘œํ˜„ ๊ณต๊ฐ„์—์„œ ๋ฐœ๊ฒฌ๋˜๋Š” ๊ณ„์ธต์  ๊ตฌ์กฐ๋ฅผ ํ†ตํ•ด ์Šคํ”ผ์ปค์˜ ์„ฑ๋ณ„, ์ถœ์‹  ์ง€์—ญ ๋“ฑ ์˜๋ฏธ๋ก ์  ์ •๋ณด๋ฅผ ์ถ”์ถœํ•  ์ˆ˜ ์žˆ์Œ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
โ€ข
HCCM ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ Liebig score๋Š” ๋ชจ๋ธ์˜ ๊ฒฐ์ • ๊ณผ์ •์„ ์ดํ•ดํ•˜๊ณ  ๊ฐœ์„ ํ•˜๋Š” ๋ฐ ํ™œ์šฉ๋  ์ˆ˜ ์žˆ๋Š” ์ƒˆ๋กœ์šด XAI ๋„๊ตฌ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
โ€ข
ํ˜„์žฌ ์—ฐ๊ตฌ๋Š” ํŠน์ • ์˜๋ฏธ๋ก ์  ํด๋ž˜์Šค์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์œผ๋ฉฐ, ๋” ๋‹ค์–‘ํ•˜๊ณ  ๋ณต์žกํ•œ ์˜๋ฏธ๋ก ์  ์ •๋ณด๋ฅผ ํฌ์ฐฉํ•˜๊ธฐ ์œ„ํ•œ ํ™•์žฅ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
๐Ÿ‘
Made with Slashpage