Coco-Nut: 自由記述文による声質制御に向けた多話者音声・声質自由記述ペアデータセット

日本音響学会2023秋季大会口頭発表資料

Transcript

1. $PDP/VU
   ࣗ༝هड़จʹΑΔ੠੍࣭ޚ ʹ޲͚ͨଟ࿩ऀԻ੠ɾ੠࣭ࣗ༝هड़
   ϖΞσʔληοτ 
   ೔ຊԻڹֶձୈճʢ೥ळقʣݚڀൃදձɹ ˑ౉ᬑ
   ဌᒀ
   ߴಓ
   ৻೭հ
   ᜊ౻
   ༎थ
   ਏ
   ℄ହ
   Ԑ౉
   ༸ ౦େӃɾ৘ใཧ޻
2. ࣗ༝هड़ʹΑΔ੠੍࣭ޚͷՄೳੑ
   w ࣗ༝౓͕ߴ͍ɼ௚ײతͳૢ࡞ΛՄೳʹ
   w େن໛ͰΦʔϓϯͳίʔύε͕ඞཁ
   Ի੠੠࣭දݱจϖΞίʔύεͷ࡞੒
   w ଟ࿩ऀԻ੠σʔλͷࣗಈऩूɾॻ͖ى͜͠
   w Ի੠΁ͷ੠࣭දݱจ෇༩ϑϩʔཱ֬
   w ໿ηάϝϯτɼ࣌ؒڧͷԻ੠ͱ੠࣭දݱจΛؚΉίʔύε͕׬੒
   ۙ೔ެ։༧ఆ ֓ཁ  ςΩετ
   Ի੠߹੒
   Ϟσϧ ੠੍࣭ޚ
   Ϟσϧ ಡΈ্͛จ ੠࣭දݱจ ࣗ༝هड़ʹΑΔ
   ੠੍࣭ޚγεςϜ ྫ
   େਓͼͨ௿͍੠Ͱམͪண͍ͯ࿩͢ɼͳͲ
3. ֓ཁ "436൛
   PO
   BS9JW
   w ରরֶशϞσϧʹΑΔఆྔධՁ͋Γ 
4. ؔ࿈ݚڀ ଟ࿩ऀɾଟελΠϧςΩετԻ੠߹੒ʹ͓͚Δ੠੍࣭ޚ
   w ࿩ऀબ୒<)PKP
   FU
   BM
   >ɼύϥϝʔλʢײ৘<-JV
   FU
   BM
   >౳ʣௐ੔
   ௿ࣗ༝౓
   w ࣗ༝هड़ʹΑΔ੠੍࣭ޚ
   <(VP
   FU
   BM
   
   :BOH
   FU
   BM
   >
   w Ի੠߹੒༻ʹऩ࿥ͨ͠গ਺࿩ऀίʔύε͕ϕʔεɾඇެ։
   ˠଟ༷ͳ࿩ऀɾ୭Ͱ΋ݚڀʹ࢖͑Δίʔύεʢ͓Αͼߏஙख๏ʣ͕ඞཁ
   5FYUUP*NBHFʹ͓͚Δଟ༷ɾެ։ίʔύεͷߩݙ
   w .4$0$0
   <-JO
   FU
   BM
   >
   w ਓखΞϊςʔγϣϯ͞ΕͨΩϟϓγϣϯը૾ϖΞίʔύε
   w ʮ
   PCKFDU
   DBUFHPSJFTʯͰʮ,ʯͷը૾ʹΩϟϓγϣϯ෇༩
   w ςΩετը૾ੜ੒ʹ͓͚Δར༻
   FH
   %"--&
   <3BNFTI
   FU
   BM
   >
   ˠଟ༷σʔλʹਓखΞϊςʔγϣϯ͞Εͨίʔύεͷެ։͕ݚڀΛଅਐ 
5. ίʔύεߏ੒ཁૉ Ի੠
   
   ॻ͖ى͜͠จʢ಺༰ϓϩϯϓτʣ
   
   ੠࣭දݱจʢ੠࣭ϓϩϯϓτʣ  த೥ͷஉੑ͕ɺ
   Θ͔Γ΍͍͢ޱௐͰ
   ஸೡʹ࿩Λ͍ͯ͠Δɻ
   ืू͍ͤͯͨͩ͘͜͞ͱͱͳ Γ·ͨ͠ɻࠓճ͸:PV5VCFͳ ͲͷϝσΟΞ࿐ग़Մೳͳʜʜ ྫ
   
6. ίʔύεߏஙख๏
7. ίʔύεߏங
   ֓ཁ  σʔλऩू ಈըϑΟϧλ ඼࣭อূ ਓखΞϊςʔγϣϯ w 8FC্ͷಈըΛࣗಈऩूʢଟ਺ͷ࿩ऀͷԻ੠Λऩूʣ
   w ੠ʹؔ܎͢ΔΩʔϫʔυͰಈըݕࡧ w Ի੠Λؚ·ͳ͍ɾԻ੠ΛؚΉ͕ಛ௃ͷബ͍੠ͷಈըΛআ֎
   w ੠ʹؔ܎͢Δࢹௌऀίϝϯτͷ਺͕ج४ w ࣗಈऩूͨ͠Ի੠ΛԻڹతɾݴޠతʹ඼࣭อূ͠ɼίʔύε ͷ࣭Λ୲อ w ऩूԻ੠ʹ੠࣭දݱจΛ෇༩
   w Ϋϥ΢υιʔγϯάΛ௨ͨ͡ෆಛఆଟ਺͔Βͷ੠࣭දݱจऩ ूϑϩʔཱ֬
8. ίʔύεߏங
   σʔλऩूɾಈըϑΟϧλ ੠ʹؔ܎͢ΔϑϨʔζΛ༻͍ͨ:PV5VCFݕࡧ
   w 8JLJQFEJB಺੠ؔ܎ΧςΰϦهࣄΛ࢖༻ͨ͠ϑϨʔζऩू
   <౉ᬑ
   FU
   BM
   >
   ੠ؔ܎ίϝϯτࣝผ
   <౉ᬑ
   FU
   BM
   >
   ʹجͮ͘੠ؔ܎ಈըநग़
   w ੠ʹؔ܎͢Δίϝϯτ͕෇͘
   
   ʢݴٴͨ͘͠ͳΔಛ௃ͷ͋ΔʣԻ੠ΛؚΉ
   w Ұఆ਺Ҏ্ͷʮ੠ʹؔ܎͢Δʯࢹௌऀίϝϯτ͕෇͍ͨಈըͷΈ࢖༻
   w ʮ੠ʹؔ܎͢Δʯίϝϯτͷࣗಈࣝผ
   <౉ᬑ
   FU
   BM
   >  ˓˓ 8JLJQFEJB $BUFHPSZ 🔍 ˚˚ ˑˑ ✅ ❌
9. ίʔύεߏங
   ඼࣭อূ Ի੠඼࣭อূ
   Իͱͯ͠ͷ඼࣭อূ
   w Ի੠۠ؒηάϝϯτݕग़ɾ෼ׂɼԻ੠ڧௐɼ௿Ի࣭Ի੠আڈɼ୹͗͢Δ௕͗͢ΔԻ੠আ ڈɼෳ਺࿩ऀԻ੠আڈɼՎ੠আڈɼ੠࣭ଟ༷ੑ࠷େԽʹΑΔൃ࿩બ୒
   w ੠࣭ଟ༷ੑ࠷େԽʹΑΔൃ࿩બ୒
   YWFDUPS
   <4OZEFS
   FU
   BM
   >
   ʹΑΔ੠࣭ϕΫτϧԽ ʹج͖ͮɼ΢Υʔυ๏ʹجͮ͘֊૚తΫϥελϦϯά
   <8BSE
   +S
   FU
   BM
   >
   ͷ͋ͱ֤Ϋϥ ελ͔ΒϥϯμϜʹൃ࿩Λબ୒
   ಺༰඼࣭อূ
   ൃ࿩಺༰ͷ඼࣭อূ
   w ॻ͖ى͜͠
   ࣗಈԻ੠ೝࣝ
   
   खಈमਖ਼
   w ඇର৅ݴޠԻ੠আڈɼෆద੾දݱʢ/4'8ʣআڈɼϊϯόʔόϧԻ੠আڈ
   Ի੠ɾ಺༰྆ଆ໘͔Βͷ඼࣭อূʹΑΓɼࣗಈऩूͨ͠Ի੠ͷ࣭Λอূ 
10. ίʔύεߏங
    ਓखΞϊςʔγϣϯ Ϋϥ΢υιʔγϯάʹΑΔ੠࣭දݱจऩू  จࣈ਺Լݶఏࣔ ྫจఏࣔ
11. ίʔύε෼ੳ
12. ίʔύε෼ੳ
    σʔλऩू֓ཁ  ݕࡧʹΑΔ֫ಘಈը਺ ໿ສ
    <౉ᬑ
    FU
    BM
    > ಈըϑΟϧλޙͷಈը਺   ηάϝϯτ਺   ߹ܭܧଓ௕ ໿࣌ؒ൒ ੠࣭දݱจ෇༩਺Ի੠ ֶशηοτ
    
    
    ݕূɼධՁηοτ
     ܧଓ௕ൣғ ඵ
13. ίʔύε෼ੳ
    ੑผͷ෼෍  0 20 40 60 Number of utterances [%] Not indicated Non-binary Female Male Male Female Non-binary Not indicated ੠࣭දݱϓϩϯϓτ಺Ͱͷʮஉʯʮঁʯͷදهʹج͖ͮ෼ྨɾूܭ
    ྆ํͷදه͕͋Ε͹/POCJOBSZɼ྆ํͳ͚Ε͹/PU
    JOEJDBUFE ੑผ͝ͱͷԻ੠ηάϝϯτ਺ ੑผ͝ͱͷYWFDUPS෼෍ʢU4/&ͰՄࢹԽʣ େ෦෼͕෼ྨՄೳ ʢҰ෦/POCJOBSZ౳ʣ உঁ͸Ϋϥελ͋Γ
    ʢ/POCJOBSZ౳͸Ϋϥελແ͘෼෍ʣ
14. ίʔύε෼ੳ
    ੠࣭දݱจྫ  ୅͘Β͍ͷஉੑͷ੠ɻΏͬ͘ΓͱԺ΍͔ͳ࿩͠ํͰͨ͠ɻۤ೰ʹຬͪ ͨɺ͚ͩΔͦ͏ͳ੠Ͱͨ͠ɻ ݩؾͳஉੑ͕໌Δ͍੠Ͱɺςϯγϣϯߴ͘ൃදΛ͢ΔΑ͏ʹ஻͍ͬͯΔɻ ए͍ঁੑ͕ɺ཈༲ͷ͋Δ੠ͰɺΏͬ͘Γͱ஻͍ͬͯΔɻ
15. ίʔύε෼ੳ
    ऩू੠࣭ プ ϩϯ プ τͷසग़୯ޠ෼ੳ
     ୯ޠͷग़ݱස౓ΛϫʔυΫϥ΢υͰਤࣔʢେ͖͘ॻ͔Εͨ୯ޠ΄Ͳසग़ʣ
    සग़୯ޠ͸සग़ଐੑɾೝࣝ͞Ε΍͍͢ཁૉΛࣔ͢ ໊ࢺ ܗ༰ࢺ ಈࢺ w :PV5VCFυϝΠϯʹ͸ૣޱͷ࿩ऀ͕ଟ͍
    w ࿩଎͸੠࣭ͷதͰ΋ೝࣝ͠΍͍͢ཁૉͰ͋Δ
16. ·ͱΊ ࣗ༝هड़Ͱ੠࣭ίϯτϩʔϧՄೳͳςΩετԻ੠߹੒ͷͨΊͷίʔύε࡞੒
    w Ի੠੠࣭ࣗ༝هड़จͷର
    w ެ։͞Ε͓ͯΓɼଟ༷ͳ੠࣭ͷ࿩ऀΛؚΈɼਓखʹΑΔʢࣗಈऩूͰ͸ͳ͘ɼ࣭͕ߴ͍ʣ ࣗ༝هड़͕෇༩͞Εͨίʔύε
    :PV5VCFͰͷԻ੠σʔλऩू
    
    ࣗಈɾਓख඼࣭อূ
    w ੠ʹؔ܎͢Δಈըͷબผ
    w Ի੠ɾ಺༰྆ଆ໘͔Βͷ඼࣭อূ
    w ਓखʹΑΔࣗ༝هड़෇༩
    w ࠓޙɼຊίʔύεΛ༻͍ɼ੠࣭දݱจͰ੠੍࣭ޚՄೳͳςΩετԻ੠߹੒ϞσϧͷߏஙΛ ߦ͏༧ఆ 
17. "QQFOEJY
    ಈըϑΟϧλͷઃఆ  ੠ؔ܎ίϝϯτ਺ᮢ஋ ݅ ΩʔϫʔυϚονϯά
    ࢖༻୯ޠ ੠ɼ ボ ΠεɼԻɼௌɼฉɼՎ λΠτϧซ༻ ͋Γ #&35ϕʔεࣝผػ࢖༻
    ୯ޠ૊Έ߹Θͤ ʮϘΠεɾฉʯʮԻʯʮ੠ʯʮԻɾฉʯ
    ʮ੠ɾϘΠεɾԻʯʮϘΠεɾԻʯʮ੠ɾԻɾՎʯ
18. "QQFOEJY
    Ի੠඼࣭อূ w Ի੠۠ؒݕग़ɾηάϝϯτநग़
    w Ի੠ڧௐ
    w ௿Ի࣭Ի੠আڈ
    w ܧଓ௕ϑΟϧλ
    w ௿ԻྔԻ੠আڈ
    w ෳ਺࿩ऀԻ੠আڈ
    w Վ੠আڈ
    w ੠࣭ଟ༷ੑ࠷େԽʹΑΔൃ࿩બ୒  
    JOB4QFFDI4FHNFOUFSʹΑΔ7"%
    
    %FNVDT
    
    /*42"ʹΑΔࣗಈ.04ධՁ
    
    dඵ
    
    ਖ਼نԽޙͷԻྔEC
    
    ਓख඼࣭อূ
    
    ਓख඼࣭อূ
    
    TDJLJUMFBSOΛ࢖༻
    
19. "QQFOEJY
    ಺༰඼࣭อূ w ඇର৅ݴޠԻ੠আڈ
    w ෆద੾දݱʢ/4'8ʣআڈ
    w ϊϯόʔόϧԻ੠আڈ  
    8IJTQFSʹΑΔ࢖༻ݴޠਪఆ
    
    
    
    ਓख඼࣭อূ
    
    /4'8ࣙॻʹΑΔΩʔϫʔυݕࡧ
    
    
    
    ਓख඼࣭อূ
    
    #&35ʹΑΔ.-.είΞධՁ
    ʢϊϯόʔόϧԻ੠͸.-.είΞ͕ ۃ୺ʹߴ͘ͳΔ܏޲͕͋Δ FH
    l͋ ͋͋͋z ʣ
20. "QQFOEJY
    ਓख඼࣭อূͷͨΊͷΫϥ΢υιʔγϯά6*  डௌධՁ࣮ݧ Ի੠Λ࠷ޙ·Ͱ࠶ੜ͠ɺ֤Ի੠͕ҎԼͷ৚݅Λຬ͍ͨͯ͠Δ͔ɺͦΕͧΕνΣοΫ͍ͯͩ͘͠͞ɻ ஻͍ͬͯΔ಺༰͕ೝࣝͰ͖ΔͳΒ͹ػցͷ੠΋ൃ࿩ऀͱͯ͠ѻ͍ͬͯͩ͘͞ɻ 1. ެংྑଏʹ൓͍ͯ͠Δ 2. ೔ຊޠ͚͕ͩ࢖ΘΕ͍ͯΔʢ೔ຊޠจதʹ୯ޠ୯ҐͰ֎ࠃޠ͕ݱΕΔ৔߹͸೔ຊޠ͚ͩͱΈͳ͢ʣ 3. ൃ࿩ऀ͸1ਓ͚ͩʢਓͷ੠͕ೖͬͨBGMɺ૬Ṁ΋ͳ͍ʣ 4. Վ͕ೖ͍ͬͯΔʢBGMͷϘʔΧϧ΋ؚΉʣ Ի੠͝ͱʹશ߲໨ʹ͍ͭͯೖྗ͕ऴΘͬͨΒ֬ఆϘλϯΛԡ͠ɺ࣍ʹਐΜͰ͍ͩ͘͞ɻ ͳ͓ɺಉ͡Ի੠͕࿈ଓͯ͠ྲྀΕΔ͜ͱ΋͋Γ·͢ɻ ࡞ۀ࣌ؒ͸3෼ఔ౓Λ༧ఆ͍ͯ͠·͢ɻ ެংྑଏʹ൓͍ͯ͠Δʁ ͸͍ ͍͍͑ ೔ຊޠ͚͕ͩ࢖ΘΕ͍ͯΔʁʢ೔ຊޠจதʹ୯ޠ୯ҐͰ֎ࠃޠ͕ݱΕΔ৔߹͸೔ຊޠ͚ͩͱΈͳ͢ʣ ͸͍ ͍͍͑ ൃ࿩ऀ͸1ਓ͚ͩʁʢਓͷ੠͕ೖͬͨBGMɺ૬Ṁ΋ͳ͍ʣ ͸͍ ͍͍͑ Վ͕ೖ͍ͬͯΔʁʢBGMͷϘʔΧϧ΋ؚΉʣ ͸͍ ͍͍͑ ֬ఆ -0:02
21. "QQFOEJY
    ਓख඼࣭อূͷͨΊͷΫϥ΢υιʔγϯάઃఆ  ϓϥοτϑΥʔϜ -BODFST
    IUUQTXXXMBODFSTKQ ධՁ͢ΔԻ੠ϑΝΠϧ਺
    ϫʔΧʔ  ༧૝ॴཁ࣌ؒ
    ϫʔΧʔ ෼ ใु
    ϫʔΧʔ =
22. "QQFOEJY
    ਓखΞϊςʔγϣϯͷͨΊͷΫϥ΢υιʔγϯάઃఆ  ϓϥοτϑΥʔϜ -BODFST
    IUUQTXXXMBODFSTKQ ධՁ͢ΔԻ੠ϑΝΠϧ਺
    ϫʔΧʔ  ༧૝ॴཁ࣌ؒ
    ϫʔΧʔ ෼ ใु
    ϫʔΧʔ =
23. "QQFOEJY
    ॻ͖ى͜͠จͷਓखमਖ਼ྫ w ࣗಈॻ͖ى͜͠͸8IJTQFSͰߦ͏
    w 5ZQPɼ8IJTQFSϞσϧߏ଄༝དྷͷϧʔϓɼ੾Ε͍ͯΔ෦෼ͷࣗಈ༧ଌ͞Εͯ ͠·ͬͨՕॴͳͲΛਓखमਖ਼  άϦʔϯं͸଍ݩͷॏྔ৐ंޱ߸ंͱ ߸ंͰ͓଴͍ͪͩ͘͞ άϦʔϯं͸଍ݩͷ྆৐ंޱ߸ं ͱ߸ंͰ͓଴͍ͪͩ͘͞
24. "QQFOEJY
    ಈըΧςΰϦͷ෼෍  0 10 20 30 40 Number of utterances [%] Sports Science & Technology Nonprofits & Activism Travel & Events News & Politics Comedy Howto & Style Music Autos & Vehicles Film & Animation People & Blogs Gaming Education Entertainment Entertainment Education Gaming People & Blogs Film & Animation Autos & Vehicles Music Howto & Style Comedy News & Politics Travel & Events Nonprofits & Activism Science & Technology Sports ಈըΧςΰϦ͝ͱͷԻ੠ηάϝϯτ਺ ಈըΧςΰϦ͝ͱͷYWFDUPS෼෍ :PV5VCFͰઃఆ͞Ε͍ͯΔಈըΧςΰϦʹج͖ͮ෼ྨ͠ूܭ
    ΧςΰϦ͝ͱͷ੠࣭ͷಛ௃ΛYWFDUPSʹΑͬͯఆྔԽ͠ɼU4/&ͰՄࢹԽ ໿ ΧςΰϦ ΫϥελʢΧςΰϦʹಛ௃తͳ੠࣭ʣ