Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Sign up for free
Menu
Search
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Pricing
Search
Sign in
Sign up for free
論文紹介:Few-shot sound event detection
Search
Sponsored
·
Your Podcast. Everywhere. Effortlessly.
Share. Educate. Inspire. Entertain. You do you. We'll handle the rest.
→
Yuya Yamamoto
May 26, 2020
340
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
論文紹介:Few-shot sound event detection
ICASSP 2020 Wangら.
Yuya Yamamoto
May 26, 2020
More Decks by Yuya Yamamoto
See All by Yuya Yamamoto
GMI44@Music and Language研究のサーベイ報告
yamathcy
1
180
APSIPA 2023 Toward Leveraging Pre-Trained Self-Supervised Frontends for Automatic Singing Voice Understanding Tasks: Three Case Studies
yamathcy
0
140
muana vol.11 音楽識別の事前学習モデル
yamathcy
1
850
国際会議ISMIR2022報告(山本分)
yamathcy
0
270
Do It Yourself: Sonic Visualiserで音楽分析してみよう
yamathcy
0
2.1k
2022年度情報学学位プログラム説明会 学生体験談
yamathcy
0
210
LSPC博士前期チュートリアル
yamathcy
0
240
MULTIMODAL METRIC LEARNING FOR TAG-BASED MUSIC RETRIEVAL@ICASSP2021読み会
yamathcy
0
1.9k
SIGMUS130-yamamoto
yamathcy
0
88
Featured
See All Featured
ReactJS: Keep Simple. Everything can be a component!
pedronauck
666
130k
Max Prin - Stacking Signals: How International SEO Comes Together (And Falls Apart)
techseoconnect
PRO
0
460
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
37
6.6k
How to Align SEO within the Product Triangle To Get Buy-In & Support - #RIMC
aleyda
2
1.8k
Product Roadmaps are Hard
iamctodd
55
13k
Neural Spatial Audio Processing for Sound Field Analysis and Control
skoyamalab
0
500
Facilitating Awesome Meetings
lara
57
7.1k
Believing is Seeing
oripsolob
1
220
Building Adaptive Systems
keathley
44
3.2k
Bash Introduction
62gerente
615
220k
10 Git Anti Patterns You Should be Aware of
lemiorhan
PRO
659
62k
Java REST API Framework Comparison - PWX 2021
mraible
34
9.7k
Transcript
'FXTIPUTPVOEFWFOUEFUFDUJPO :V8BOH +VTUJO4BMBNPO /JDIPMBT+#SZBO +VBO1BCMP#FMMP /FX:PSL6OJWFSTJUZ "EPCF3FTFBSDI *$"441 αʔϕΠൃද ஜେֶ
.ࢁຊ ༤
ͦͷલʹ • *$"441ɿԻڹ৴߸ॲཧͷτοϓΧϯϑΝϨϯε • Ի͚ͩͰͳ͘ը૾ηϯασʔλ৴߸ॲཧத৺ʹϨϕϧͷߴ͍ݚڀ ͕ग़ଗ͏ • ࠓճεϖΠϯͷόϧηϩφ͕ͩͬͨίϩφͷӨڹͰόʔνϟϧ։࠵ 2
*$"441Ͱ͕ײͨ͡τϨϯυ • ੜܥˠΑΓࣗવԻʹۙ͘ɼ͋Δ͍ίϯτϩʔϥϒϧʹԻΛੜ͢Δ • ಛྔͷղ͖΄͙͠දݱɿʮ%JTFOUBOHMFʯͱ͍͏ΩʔϫʔυͰώοτ͠·͢ • 8BWF/FUɾ7"&ɾ("/ͳͲΓΓ • %FFQ3FE0DFBOͩͳ͋ খฒײ
• ೝࣝܥˠΑΓগͳ͍ϥϕϧσʔλͰͳΜͱ͔͢Δ • ڭࢣ͋Γֶशɿϥϕϧͷ͋Δσʔλͱϥϕϧͷͳ͍σʔλΛֶࠞͥͯश • /4IPUֶशɿݕग़ର͕͘͝গ͔͠ूΊΒΕͳ͍߹ • 4FMG4VQFSWJTJPOɿڭࢣͳ͠σʔλ͔ΒݸʑΛݟ͚ΔಛΛൃݟ͢Δ 3
*$"441Ͱײͨ͡τϨϯυ • ઃఆ͕໘ന͍ͷ࣮ར͕େ͖͍ɼϨουΦʔγϟϯʹߦ͔ͳ͔ͬͨ ݚڀগͰ͋Δ͕ଘࡏ • ࠓͷྫ • ՎࢺͷೝࣝɾΞϥΠϝϯτΛ͚ͭͨ΄͏͕͏·͍͘͘ͷͰʁΛݕূ <(VQUB> •
ਓؒΛ("/ͷ%JTDSJNJOBUPSͱͯ͠͏Ի߹ <'VKJJ > • ָۂ͔ΒG Lɼ&YQMJDJUͳՎࢺΛݕग़ <7BHMJP > • ͍ͣΕʹͤΑɼ*4.*3ΑΓԥΓ߹͍ײ͕ڧ͍ձٞͩͳ͋ͱײͨ͡ 4
͜ͷจ͕ͬͨ͜ͱ • গྔͷֶशσʔλͷΈͰԻ͔ΒಛఆͷΠϕϯτΛݕग़͢Δ 'FXTIPUԻڹΠϕϯτݕग़ 5
ͱͱͷϞνϕʔγϣϯ φϨʔγϣϯಈըɾԻฤूͰ༨ܭͳ෦ΛऔΓআ͖͍ͨ ݴ͍ཎΈɼ֏͍ɼטΉɼࡶԻ͕ೖΔ ˠ͍͟ฤूͯ͠ΈΔͱɼऔΓআ͘ͷେมʂ ˠಉ͡Α͏ͳϛεςΠΫͳΒࣗಈݕग़ͯ͠ҰؾʹΧοτͰ͖ͳ͍͔ʁ 6 ࠓ ͑ʔ θϛΛ
͑ʔΖ͏ͱ͑ʔ ࢥͬͯ͑ʔ ߦ͜͏ͱ͑ʔ 「えー」って⾔いすぎじゃね? カットしたいけど全部⼿作業で トリミングすんのめんどくさ... 1個めの「えー」を選択するだけで 録⾳中の「えー」全部消せないかなぁ
͜ͷΛղܾ͢ΔͨΊͷোน ΧοτରͱͳΔҟԻΛݕग़͍ͨ͠ ˠ΄ͱΜͲͷ߹σʔληοτΛूΊΔͷෆՄೳ ݕग़ରཏສͷԻͷதͰۃΊͯগ ֶशσʔληοτʹͳ͘ɼ৽ͨʹΛूΊΒΕͳ͍߹ଟ͍ 7
ղܾࡦ 'FX4IPU-FBSOJOH • গͳϥϕϧ͖ͭσʔλΛݕग़Ͱ͖ΔΑ͏ʹ͢Δֶशํ๏ • ௨ৗͷڭࢣ͋ΓֶशͱҟͳΓɼϥϕϧʹؔ͢ΔͳΜΒ͔ͷิॿใ Λར༻ͯ͠ݕग़Λߦ͏ • ػցֶशʹඞཁͳσʔλͷ ϥϕϧ༩ʹؔΘΔࠔΛ
ղܾ͢Δख๏ͱͯ͠ ͕ू·͍ͬͯΔ 8
'FXTIPUMFBSOJOH͍Ζ͍Ζ ϥϕϧใͷηϚϯςΟΫεΛར༻ • ϥϕϧͷݴޠใΛࢄදݱʹͯ͠ڞʹۭؒຒΊࠐΈ <'SPNF > • ϥϕϧࣗମͷ֊ߏΛར༻<-J> 9 ڑֶशΛར༻
ʮͦͦಉ͡Α͏ͳͷΛ͚ۙͮΔΑ͏ ͳۭؒࣗମΛ࡞Γग़ͯ͠͠·͓͏ʯ ɾLNFBOTతͳηϯτϩΠυΛ༻ҙ͢Δ 1SPUPUZQJDBM/FUXPSL<4OFMM> ΄͔͍Ζ͍Ζ ࠓճͬͪ͜
'FXTIPU-FBSOJOH • $XBZ,TIPUྨ • $छྨͷগͷϥϕϧ͖ͭσʔλ͕,ݸ͋Δ߹ͷྨ • ͜ͷྫͩͱछྨͷϥϕϧɾݸͷσʔλ͕͋ΔͷͰXBZTIPU 10 サポートセット: 少数しかない検出対象を
かき集めた 学習データのようなもの クエリ: 検出対象となる⼊⼒
ڑֶशϕʔε'FXTIPU-FBSOJOH • ຒΊࠐΈ෦ɿσʔλΛϕΫτϧԽͯ͠ಉۭؒ͡ʹຒΊࠐΈ • ྨࣅఆ෦ɿΫΤϦʢఆ͍ͨ͠σʔλʣ͕ͲΕʹ͍͔ۙܭࢉ 11 CNNが エンコーダとして よく⽤いられる コサイン類似度や
ユークリッド距離, 類似度算出のための ネットワークなど G_simでの 類似計算の結果, 最も近いラベルに分類
࣮ࡍͲ͏ݕग़͢Δͷ͔ • αϙʔτηοτͱͯ͠ݕग़ରͷԻΛ͍͔ͭ͘બͯ͠ਖ਼ྫͱ͢ΔʢDʣ • NTͣΒͯ͠σʔλ֦ுͰ͖Δ • ͦΕҎ֎ͷదͳԻෛྫͱͯ͠ѻ͍ɼαϙʔτηοτʹϥϯμϜબग़ ʢݕग़ରͦ͏ଟ͘ͳ͍ͱԾఆʣʢCʣ • ΓͷϑϨʔϜΫΤϦͱͯ͠ݕग़ʹ༻
12
ωοτϫʔΫΛֶश͢Δํ๏ &QJTPEJD5SBJOJOH • େنͳσʔληοτͰϥϕϧΛछྨબͼɼ'FXTIPUMFBSOJOHΛ ࣮ࢪˠϥϕϧͷΈ߹ΘͤΛม͑Δ • ϞσϧࣝผͷํΛֶश͠ɼঃʑʹΫϥεඇґଘͷࣝผೳྗΛ֫ಘ • ྨϩεΛ࠷খԽ͢ΔΑ͏ֶशͤ͞Δ 13
࣮ݧ • Ի͔Βͷϫʔυݕग़ • ݕग़͍ͨ͠ରΛ͍͔ͭ͘બΜͰαϙʔτηοτͱͯ͠༩͑ɼ ΫΤϦҎ֎ͷԻΫϦοϓ͔Β୯ޠΛݕग़͢Δ 14 ͑ʔ θϛΛ͍·͔ΒΖ͏ͱ ͑ʔ
ࢥ͍ͬͯ·͢ ͚Ͳ ͑ʔ ࠓ
ݕূ͢Δࣄ߲ • ෳ͋Δྨࣅࢉग़Ͳͷํ๏͕Α͍͔ʁ • ਖ਼ྫͱෛྫΛͲͷ͘Β͍αϙʔτηοτʹूΊΕΑ͍͔ʁ • ʢυϝΠϯͷҟͳΔσʔληοτʹରͯ͠ػೳ͢Δ͔ʁʣ 15
࣮ݧɿͲͷྨࣅࢉग़ख๏͕͍͍͔ 'FX4IPU-FBSOJOHͷख๏ΛऔΓ্͛ͯੑೳΛࢼ͢ • 4JBNFTF/FUXPSL<,PDI>ˠͭͷೖྗ͕ಉ͡ͳΒ-ڑΛۙ͘ • .BUDIJOH/FUXPSL<7JOZBMT >ˠ֤ϥϕϧͱͷ$PTڑΛۙ͘ • 1SPUPUZQJDBM/FUXPSL<4OFMM>ˠ֤ϥϕϧͷฏۉϕΫτϧͱϢʔΫϦουڑΛۙ͘ •
3FMBUJPO/FUXPSL<4VOH>ˠ$//Ͱྨࣅܭࢉ ˠओͳҧ͍ྨࣅఆ෦ɽಉ͡ϥϕϧͷϕΫτϧ͕ۙ͘͢ΔΑ͏ʹ ຒΊࠐΈۭؒΛֶश͢Δͱ͍͏ΞΠσΞಉ͡ 16
֤ख๏ͷਤ 17
࣮ݧɿ֤छ࣮ݧ݅ • σʔληοτ • 4QPLFO8JLJQFEJB$PSQPSB<,PIO> • 8JLJQFEJBͷӳޠ࿕ಡσʔληοτˠϫʔυΛݕग़ରͱ͢Δ • ೖྗ •
ඵͷԻΛL)[ͰαϯϓϦϯά • NT૭෯ɼNTͣΒ͠෯ɼ࣍ݩͷϝϧεϖΫτϩάϥϜʹม ˠ º ͷߦྻ • ຒΊࠐΈ෦ • ͷ$//ΤϯίʔμʢºϑΟϧλɼº.BYQPPMɼ׆ੑԽ 3F-6ʣ • ྨࣅఆ෦ • ֤ख๏ͰҟͳΔ 18
࣮ݧɿධՁج४ • "613$ʀ"WFSBHFBSFB6OEFSUIF1SFDJTJPO3FDBMM$VSWF • "6$30$ͷ13ۂઢόʔδϣϯ • σʔλ͕*NCBMBODFʢ͕ෛྫਖ਼ྫͳ߹ͱ͔ʣʹΑ͘༻͍ΒΕΔ • ݕग़ΛऔΓ͜΅ͣ͞ɼ͔ͭਖ਼֬ʹͯΒΕ͔ͨΛධՁ 19
IUUQTXXXZPVUVCFDPN XBUDI W))KXP$U:" ΑΓҾ༻
݁ՌɿͲͷྨࣅࢉग़͕͍͍͔ • $ɿϥϕϧͷछྨɼ ,ɿֶशͰ͖ΔαϯϓϧΛมԽͤ͞ ࣮ݧ • Qਖ਼ྫͷ • 1SPUPUZQJDBM/FUXPSL͕ଞΑΓ ൺֱతߴ͍"613$ΛϚʔΫ
20
݁ՌɿෛྫΛมԽͤͨ࣌͞ͷύϑΥʔϚϯεਪҠ • ෛྫͷΛOͱ͠ɼ $ , Q
ɼσʔλ֦ுͳ͠Ͱݻఆ • OͰɼΫϦοϓதਖ਼ྫҎ֎ͷશͯΛͬͨ߹ͱఔͷޡࠩ ˠෛྫਖ਼ྫͷഒఔ͋ΕΑ͍ʁ 21 Allと1%以下の誤差 である74.69%
݁Ռɿਖ਼ྫΛมԽͤͨ࣌͞ͷύϑΥʔϚϯεਪҠ • ਖ਼ྫͷΛQͱ͢Δ • Yʢ֦ுʣ ͰมԽͤ͞Δ • ʢ$ ,
Oʣʢ BMMʣͰݻఆͨ͠ͱ͜Ζɼ QͰͷਫ਼ ˠݸબ͢Ε͘Β͍ݕग़Ͱ͖Δ • σʔλ֦ுͷޮՌബ͍ • ݕग़ରͷXPSE͕ඵΑΓ͍ͱμϝʁ ˠݕग़ରΛdඵͷXPSEʹݶఆ͢ ΔͱQ ͷ߹ͷ݁Ռ͕ಛʹվળ 22
͓·͚݁ՌɿӳޠͰֶशͤͨ͞ϞσϧଞݴޠͰػೳ͢Δʁ • Φϥϯμޠ %VUDI υΠπޠ (FSNBO Ͱӳޠͱಉͷ ੑೳΛϚʔΫ • ࣮ݧͰͬͯͳ͔͚ͬͨͲྲྀੴʹޠ͕ΕͯΔͱແཧ͔ʁ
23
·ͱΊ • ઓͨ͠ɿ໘ͳɼϑΝΠϧதͷࣅͨԻ୳͠ΛࣗಈͰ • ղܾࡦɿڑֶशΛجʹͨ͠'FXTIPU-FBSOJOH • ํ๏ɿ • ֶशɿ&QJTPEJDUSBJOJOHͰϞσϧΛֶश •
ςετɿݕग़ର͔൱͔Λྨ͢ΔɼෛྫϥϯμϜʹબग़ • ݁Ռɿ • QSPUPUZQJDBMOFUXPSL͕࣮ݧతʹΑ͍͜ͱΛ֬ೝ • ਖ਼ྫݸͷબͰͷੑೳΛൃش • ֶश࣌ͷΫϦοϓͷ͞ʹऩ·Δݕग़ରਖ਼ྫݸͷબͰͷੑೳ • ӳޠͷձͰֶशͨ͠ϞσϧΦϥϯμޠɼυΠπޠͰػೳ 24
"EPCFͷެࣜൃදσϞ • IUUQTZPVUVCFFC"MO89TL 25
ࢀߟจݙ • <8BOH>8BOH : 4BMBNPO + #SZBO /+ #FMMP +1
.BZ 'FX4IPU4PVOE&WFOU%FUFDUJPO*O*$"441*&&&*OUFSOBUJPOBM$POGFSFODFPO"DPVTUJDT 4QFFDIBOE4JHOBM1SPDFTTJOH *$"441 QQ *&&& • <(VQUB>(VQUB $ :MNB[ & -J ) .BZ "VUPNBUJD-ZSJDT"MJHONFOUBOE5SBOTDSJQUJPOJO1PMZQIPOJD.VTJD%PFT#BDLHSPVOE.VTJD)FMQ *O *$"441 *&&&*OUFSOBUJPOBM$POGFSFODFPO"DPVTUJDT 4QFFDIBOE4JHOBM1SPDFTTJOH *$"441 QQ *&&& • <'VKJJ >'VKJJ , 4BJUP : 5BLBNJDIJ 4 #BCB : 4BSVXBUBSJ ) .BZ )VNBO("/HFOFSBUJWFBEWFSTBSJBMOFUXPSLXJUIIVNBOCBTFEEJTDSJNJOBUPSBOEJUTFWBMVBUJPOJO TQFFDIQFSDFQUJPONPEFMJOH*O*$"441*&&&*OUFSOBUJPOBM$POGFSFODFPO"DPVTUJDT 4QFFDIBOE4JHOBM1SPDFTTJOH *$"441 QQ *&&& • <7BHMJP >7BHMJP " )FOOFRVJO 3 .PVTTBMMBN . 3JDIBSE ( E`"MDIÉ#VD ' .BZ "VEJP#BTFE%FUFDUJPOPG&YQMJDJU$POUFOUJO.VTJD*O *$"441*&&& *OUFSOBUJPOBM$POGFSFODFPO"DPVTUJDT 4QFFDIBOE4JHOBM1SPDFTTJOH *$"441 QQ *&&& • <'SPNF >'SPNF " $PSSBEP (4 4IMFOT + #FOHJP 4 %FBO + 3BO[BUP ." .JLPMPW 5 %FWJTF"EFFQWJTVBMTFNBOUJDFNCFEEJOHNPEFM*O "EWBODFTJOOFVSBM JOGPSNBUJPOQSPDFTTJOHTZTUFNT QQ • <-J>-J " -VP 5 -V ; 9JBOH 5 8BOH - -BSHFTDBMFGFXTIPUMFBSOJOH,OPXMFEHFUSBOTGFSXJUIDMBTTIJFSBSDIZ*O1SPDFFEJOHTPGUIF*&&&$POGFSFODFPO $PNQVUFS7JTJPOBOE1BUUFSO3FDPHOJUJPO QQ • <,PDI>,PDI ( ;FNFM 3 4BMBLIVUEJOPW 3 +VMZ 4JBNFTFOFVSBMOFUXPSLTGPSPOFTIPUJNBHFSFDPHOJUJPO*O*$.-EFFQMFBSOJOHXPSLTIPQ 7PM • <7JOZBMT >7JOZBMT 0 #MVOEFMM $ -JMMJDSBQ 5 8JFSTUSB % .BUDIJOHOFUXPSLTGPSPOFTIPUMFBSOJOH*O"EWBODFTJOOFVSBMJOGPSNBUJPOQSPDFTTJOHTZTUFNT QQ • <4OFMM>4OFMM + 4XFSTLZ , ;FNFM 3 1SPUPUZQJDBMOFUXPSLTGPSGFXTIPUMFBSOJOH*O"EWBODFTJOOFVSBMJOGPSNBUJPOQSPDFTTJOHTZTUFNT QQ • <4VOH>4VOH ' :BOH : ;IBOH - 9JBOH 5 5PSS 1) )PTQFEBMFT 5. -FBSOJOHUPDPNQBSF3FMBUJPOOFUXPSLGPSGFXTIPUMFBSOJOH*O1SPDFFEJOHTPGUIF*&&& $POGFSFODFPO$PNQVUFS7JTJPOBOE1BUUFSO3FDPHOJUJPO QQ • <,PIO>",PIO '4UFHFO BOE5#BVNBOO .JOJOHUIFTQPLFOXJLJQFEJB GPSTQFFDIEBUBBOECFZPOEJO1SPDFFEJOHTPGUIF5FOUI*OUFSOBUJPOBM$POGFSFODFPO-BOHVBHF 3FTPVSDFTBOE&WBMVBUJPO -3&$ 1BSJT 'SBODF .BZ 27