Upgrade to Pro — share decks privately, control downloads, hide ads and more …

#muana IRM

Ae87a6075a4bc957a6775c6d70c8be90?s=47 Yuya Yamamoto
February 26, 2021

#muana IRM

2021/02/26で発表したIRM×歌いにくさについて 

Ae87a6075a4bc957a6775c6d70c8be90?s=128

Yuya Yamamoto

February 26, 2021
Tweet

Transcript

  1. ࠷ۙͷώοτιϯά͸ Վ͍ʹ͍͘ͷ͔ʁ ʙೝ஌తԻָཧ࿦Ͱ෼ੳͯ͠Έͨʙ Music×Analytics Meetup #4 ࢁຊ ༤໵ (@yamathcy) 2021/02/26

  2. ࣗݾ঺հ • ࢁຊ ༤໵ʢ΍·΋ͱ Ώ͏΍ʣ@yamatchy • ஜ೾େֶେֶӃਤॻؗ৘ใϝσΟΞݚڀՊത࢜લظ՝ఔ2೥ • य़͔Βത࢜ޙظ՝ఔʹਐֶ༧ఆ •

    ݚڀ෼໺ɿԻָ৘ใॲཧʢಛʹՎʣ • ݚڀͷϞνϕɿΧϥΦέͷेീ൪୳͠Λͯ͘͠ΕΔγεςϜ͕ཉ͍͠ʂ • ݚڀςʔϚɿ • ਓ͕ײ͡ΔՎ͍ʹ͘͞ͷఆྔԽʢֶ෦ʣ • ػցֶशɾਂ૚ֶशʹΑΔՎএςΫχοΫͷ෼ੳʢӃʣ • ޷͖ͳΞʔςΟετɿΰʔϧσϯϘϯόʔ
  3. ࠓ೔ͷ಺༰ • ਓͷೝ஌Λߟྀͨ͠Իָཧ࿦Ͱ͋Δɼೝ஌తԻָཧ࿦ͷҰͭ ҉ҙ࣮ݱϞσϧΛ࢖ͬͯՎͷෳࡶ͞ΛϞσϧԽ͓ͨ͠࿩

  4. ಥવͰ͕͢ • ࠷ۙਓؾͷۂɼՎ͏ͷ೉͍͠ۂଟ͘ͳ͍Ͱ͔͢ʁʁ • ۂΛ࡞ͬͨຊਓͰ͢Β೉͍͠ͱݴ͍ͬͯΔέʔε΋…

  5. Վ͏ͷ͕೉͍͠ϝϩσΟʁ • ͍ΖΜͳཁૉ͕͋Δ • ԻҬɿࣗ෼ͷ੠ҬΛ֎ΕΔͱ೉͍͠ • ςϯϙɿ଎͍ͱ೉͍͠ • ෳࡶ͞ɿෳࡶͩͱ೉͍͠ •

    ՎࢺͷൃԻͷ͠ʹ͘͞ɿטΜ͡Ό͏ • etc… Իߴ #1.ͱԻՁ ՎࢺͷԻӆ ʁ
  6. ෳࡶ͞ʁ • ෳࡶͳϝϩσΟ #ͱ͸ →ʢೝ஌తʹʣԻΛ౰ͯͮΒ͍ϝϩσΟʁ →࣍ͷԻΛ༧ଌͮ͠Β͍Α͏ͳԻͷฒͼʁ ༧૝௨Γ😄 ༧૝֎😳

  7. ҉ҙ࣮ݱϞσϧ (Implication-Realization Model) • ೝ஌తԻָཧ࿦ͷҰͭɼϝϩσΟʹର͢Δཧ࿦ • ʮਓ͸࣍ʹԿ͕ى͜Δ͔༧ଌ͠ͳ͕ΒԻָΛௌ͘ʯ • ϝϩσΟΛʮ҉ҙʯʹج͍ͮͯͦͷ෦෼ྻΛγϯϘϧԽ •

    ҉ҙɿ2Ի࿈ଓͯ͠ௌ͍ͨ࣌ɼͦͷ࣍ͷԻΛਓ͕Ͳ͏༧૝͢Δ͔ • ࣮ݱɿ҉ҙͷ௨Γʹͳ͔ͬͨͲ͏͔Ͱ෼ྨ Ի໨ͱԻ໨ͷ͕ࠩ Ͳ͏ͳ͍ͬͯΔ͔ Ի໨ͱԻ໨ͷࠩ͸ʁ ʢԻఔݪཧʣ ˠͱˠͷํ޲͸ಉ͡ʁ ʢํ޲ݪཧʣ ڱ͍ ڱ͍ ಉ͡ํ޲ ޿͍ ڱ͍ ҟͳΔํ޲ Eugene Narmour, “The Analysis and Cognition of Melodic Complexity”, The university of Chicago press ,1992
  8. ਤղ FYڱ͍Իఔ ʁ ࣍͸zڱΊzʹ l্͕Δzʁ 🤔 ҉ҙ 3Ի໨ ҉ҙ͕࣮ݱ ->γϯϘϧ”P”

    ҉ҙͷ͏ͪ ํ޲ݪཧ͕൱ఆ ->γϯϘϧ”IP” ҉ҙͷ͏ͪ Իఔݪཧ͕൱ఆ ->γϯϘϧ”VP” ڱΊʹ ্͕Δ ڱΊʹ Լ͕Δ ޿Ίʹ ্͕Δ
  9. ਤղ P γϯϘϧɿ

  10. ਤղ P γϯϘϧɿ P

  11. ਤղ P γϯϘϧɿ P ID

  12. ਤղ P γϯϘϧɿ P ID P

  13. ਤղ P γϯϘϧɿ P ID P P

  14. ਤղ P γϯϘϧɿ P ID P P γϯϘϧͷׂ߹ɹPɿ80ˋ ID: 20ˋ

  15. ҉ҙ࣮ݱϞσϧͷγϯϘϧ ҉ҙͷݪཧͷ࣮ݱʹج͍ͮͯ෼ྨ جຊͷγϯϘϧ ྫ֎ͷγϯϘϧ 2ԻͰ׬݁ 1ԻͰ׬݁ rP rIP ಉํ޲Ͱ޿͍·· ٯํ޲

    ڱ->޿
  16. ҉ҙͷݪཧͱجຊγϯϘϧͷؔ܎ γϯϘϧ ԻఔΠϝʔδ ਐߦํ޲ Իߴࠩ Իఔݪཧ ํ޲ݪཧ 1 PS ಉํ޲

    ڱˠڱ ˓ ˓ *1 PS ٯํ޲ ڱˠڱ ˓ º 71 PS ಉํ޲ ڱˠ޿ º ˓ 3 PS ٯํ޲ ޿ˠڱ ˓ ˓ *3 PS ಉํ޲ ޿ˠڱ ˓ º 73 PS ٯํ޲ ޿ˠ޿ º ˓ % ෆม ͳ͠   *% PS ٯํ޲ ڱˠڱ Ի໨ͷԻߴ ʹԻ໨ͷԻߴ ˓ º
  17. ҉ҙͷݪཧͱྫ֎γϯϘϧͷؔ܎ γϯϘϧ ԻఔΠϝʔδ ਐߦํ޲ Իߴࠩ Իఔݪཧ ํ޲ݪཧ S1 PS ಉํ޲

    ޿ˠ޿ º º S*1 PS ٯํ޲ ڱˠ޿ º º %ZBE  ԻͷΈ͕ ݽཱ    .POBE  ԻͷΈ͕ ݽཱ   
  18. • ಐགɾՎ͍΍͍͢ۂɾՎ͍ʹ͍͘ۂͰ҉ҙ࣮ݱϞσϧͷɹɹɹɹɹ γϯϘϧग़ݱ౓߹͍ͷҧ͍͸ݟΒΕΔʁ ࠓճߦ͏෼ੳ

  19. ෼ੳָۂ • 3܈ʹ෼͚Δ ಐག ؆୯Ί ෳࡶΊ ͲΜ͙Γ͜Ζ͜Ζ ͪΐ͏ͪΐ ϩϯυϯڮ ੈքʹҰ͚ͭͩͷՖ

    4."1 νΣϦʔεϐοπ ϚϦʔΰʔϧυ͍͋ ΈΐΜ ՖՐBJLP -FNPOถ௡ݰࢣ ໷ʹۦ͚Δ:0"40#* ന೔,JOH,OV
  20. ෼ੳख๏ 1. ָ֤ۂͷϝϩσΟϥΠϯͷ1൪ͷΈͷMIDIΛೖྗ 2. ϝϩσΟͷઌ಄͔Β҉ҙ࣮ݱϞσϧʹΑΔ෼ੳΛߦ͏ 3. 2෼ԻූҎ্ͷ௕͞ͷԻූͰϑϨʔζΛ۠੾Δ 4. 3.Ͱ༩͑ͨϑϨʔζʹॱ࣍γϯϘϧ෇༩ 5.

    γϯϘϧͷׂ߹Λࢉग़
  21. ݁Ռɿಐག M 0% Dyad 0% D 0% P 41% R

    0% ID 32% IP 27% VP 0% IR 0% VR 0% rP 0% rIP 0% E 0% ロンドン橋 M Dyad D P R ID IP VP IR M 0% Dyad 0% D 5% P 12% R 8% ID 7% IP 61% VP 0% IR 0% VR 0% rP 0% rIP 7% E 0% どんぐりころころ M Dyad D P R ID IP VP IR Ի਺͕গͳ͍ͷͰׂ߹͕ૈΊ ϩϯυϯڮ͸3छྨ M 0% Dyad 0% D 24% P 13% R 0% ID 4% IP 59% VP 0% IR 0% VR 0% rP 0% rIP 0% E 0% ちょうちょ M Dyad D P R ID IP VP IR VR
  22. ݁Ռɿ؆୯Ί M 0% Dyad 0% D 29% P 17% R

    3% ID 5% IP 43% VP 0% IR 0% VR 0% rP 0% rIP 3% E 0% 世界に⼀つだけの花/SMAP M Dyad D P R ID IP VP IR M 0% Dyad 1% D 14% P 24% R 2% ID 15% IP 42% VP 0% IR 0% VR 0% rP 0% rIP 2% E 0% マリーゴールド/あいみょん M Dyad D P R ID IP VP IR VR M 0% Dyad 1% D 10% P 33% R 1% ID 12% IP 41% VP 1% IR 0% VR 0% rP 0% rIP 1% E 0% チェリー/スピッツ M Dyad D P R ID IP VP IR VR rP rIP ಐགʹൺ΂ͯภΓ͕ݟΒΕͨ ->ϑϨʔζͷόϦΤʔγϣϯ͕ଟ͍ʁ
  23. ݁Ռɿ೉͠Ί M 0% Dyad 1% D 26% P 12% R

    3% ID 12% IP 43% VP 1% IR 0% VR 0% rP 0% rIP 2% E 0% 花⽕/AIKO M Dyad D P R ID IP VP IR VR rP rIP M 0% Dyad 1% D 7% P 19% R 5% ID 14% IP 49% VP 1% IR 0% VR 0% rP 0% rIP 4% E 0% ⽩⽇/KING GNU M Dyad D P R ID IP VP IR VR rP M 0% Dyad 8% D 3% P 40% R 4% ID 16% IP 23% VP 1% IR 1% VR 0% rP 0% rIP 4% E 0% 夜に駆ける/YOASOBI M Dyad D P R ID IP VP IR VR rP rIP M 0% Dyad 0% D 3% P 43% R 5% ID 18% IP 26% VP 2% IR 0% VR 0% rP 0% rIP 3% E 0% LEMON/⽶津⽞師 M Dyad D P R ID IP VP IR VR rP ؆୯Ί܈ʹൺ΂ͯID,R͕ଟΊɼग़ݱγϯϘϧ਺΋ଟΊ ্ԼͰෳࡶ͞ͷੑ࣭͕ҟͳΔʁ
  24. ݁࿦ • ҉ҙ࣮ݱϞσϧʹΑΔ෼ੳʹΑΕ͹… • ಐགΑΓ΋γϯϘϧʹόϦΤʔγϣϯ͋Γ • ࠷ۙͷۂ͸एׯIDͱRͷγϯϘϧͷׂ߹͕ଟ͍ • ෳࡶ͞ʹ͸ɼ·͍͔ͨͭ͘ͷύλʔϯ͕͋Γͦ͏ •

    ΑΓଟ͘ͷۂͷ෼ੳ͕ඞཁ ຊ౰͸౷ܭతԾઆݕఆ/෼ࢄ෼ੳ·Ͱ΍ͬͯ ༗ҙʹ͕ࠩ͋Δ͔Λݟ͔͕ͨͬͨ ؒʹ߹͍·ͤΜͰͨ͠…
  25. Thanks!! • ҉ҙ࣮ݱϞσϧؔ࿈ • 2020೥౓ஜ೾େֶฏլݚڀࣨଔۀͷઌഐํ • ஑ాपฏ͞ΜʢγϯϘϧ෇༩ϓϩάϥϜ࣮૷/JAZZ΁ͷద༻ʣ • ٶᖒڹ͞ΜʢϙϐϡϥʔԻָ΁ͷద༻ʣ •

    ັྗతͳϝϩσΟɾՎΛಧ͚͍ͯͩͬͯ͘͞Δɹɹɹɹ ΞʔςΟετͷօ༷ • ʢՎ͍ʹ͍͘ͱ͔disͬΆ͍͜ͱݴͬͯ͢Έ·ͤΜͰͨ͠… ͜͏͍͏ϝϩσΟ΋ັྗ ʹᷓΕ͍ͯͯେ޷͖Ͱ͢ʣ
  26. ิॿࢿྉ ʴ ͓·͚

  27. ࠓޙ • ෳࡶ͞͸Ͳ͔͜ΒʁΛ΋ͬͱৄ͘͠෼ੳ͍ͨ͠ • ίʔυਐߦͱϝϩσΟͷ૊Έ߹Θͤ • ͨ͘͞ΜςϯγϣϯίʔυΛ࢖͏ͱ೉͍͠ • ϝϩσΟ͕ϧʔτԻͰͳ͍ͱෳࡶʹײ͡Δ •

    ୯Ұίʔυ಺ͰԻ͕ಈ͘ͱෳࡶʹײ͡Δ • ϦζϜ+Իߴύλʔϯͷ෼ੳ • ϝϩσΟ͕ϧʔτԻͰͳ͍ͱෳࡶʹײ͡Δ • ୯Ұίʔυ಺ͰԻ͕ಈ͘ͱෳࡶʹײ͡Δ
  28. ϝϩσΟϥΠϯͷ೉͠͞Λ෼ੳ͢Δ • Իָ৘ใՊֶݚڀձͰ༧උ୳ࡧωλͱͯ͠౤ߘ • Իߴ΍BPM౳ͷಛ௃ɾ౷ܭྔ+೉қ౓ධՁͱ૬ؔ෼ੳ΄͔ • Վ͍ʹ͘͞ͷܥ౷Λେ·͔ʹ௫Ή΋ɼݸਓ͕ࠩେ͖͘ɼཁҼ΋๲େͰ͋ΔͨΊ खʹෛ͓͑ͣଂೖΓঢ়ଶ • ༷ʑͳಛ௃Λߟ͑Δ΋݁ہɼෳࡶ͞

    is Կʁͱ͍͏ϞϠϞϠEndʹ ࢁຊ༤໵ɼฏլৡɿՎ͍΍͢͞ɾՎ͍ʹ͘͞ʹண໨ָͨ͠ۂݕࡧγεςϜͷͨΊͷϙϐϡϥʔָۂͷՎএ೉қ౓ࢉग़ͷݕ౼ɼ৘ใॲཧֶձݚڀใࠂɼVol.2019‐MUS‐ 124ɼNo.9 (2019)
  29. Q. σΟʔϓϥʔχϯάͰͳΜͱ͔ͳΒͳ͍ͷʁ • A. ଟ෼ͳΓ·ͤΜ • σΟʔϓϥʔχϯά͸ϥϕϧͱσʔλ͕ͨ͘͞Μඞཁ • ָේσʔλࣗମ͕खʹೖΓʹ͍͘ •

    Վ͍ʹ͘͞ʹؔ͢Δϥϕϧ৘ใ͸΋ͬͱखʹೖΓʹ͍͘ • ࠷ۙ͸ϝλσʔλ͕ͳͯ͘΋ಛ௃Λ΋ͱʹ෼͚ͯ͘ΕΔɼɹɹɹɹ ਂ૚ڭࢣͳֶ͠श͕τϨϯυͷҰͭɽଟগ͸ଠ౛ଧͪͰ͖ΔΑ͏ʹ ͳΔʁ
  30. ҉ҙ࣮ݱϞσϧͷݚڀࣄྫ • ϝϩσΟੜ੒ • Yazawa, S., Hamanaka, M., & Utsuro,

    T. (2014, August). Melody generation system based on a theory of melody sequence. In 2014 International Conference of Advanced Informatics: Concept, Theory and Application (ICAICTA) (pp. 336-341). IEEE. • ϑϨʔζͷ෼ࢄදݱͷධՁʹ༻͍ͨ • Hirai, T., & Sawada, S. (2019). Melody2vec: Distributed representations of melodic phrases based on melody segmentation. Journal of Information Processing, 27, 278-286. • ༷ʑͳδϟϯϧͷ෼ੳ΁ͷԠ༻ • ೳొ෨, ஛઒Ղ੒, & ฏాܓೋ. (2018). ҉ҙ࣮ݱϞσϧʹج͖ͮ࡞ۂՈࣝผΛߦ͏ HMM ʹ͍ͭͯ. ݚڀใࠂԻ੠ݴޠ৘ใॲཧ (SLP), 2018(18), 1-6. • ٶᖒڹ, & ฏլৡ. (2019). ҉ҙ࣮ݱϞσϧʹجͮ͘ϙϐϡϥʔԻָͷηΫγϣϯ෼ ੳ. ୈ 81 ճશࠃେձߨԋ࿦จू, 2019(1), 371-372.
  31. ͓·͚ᶄɿԻָ৘ใॲཧͰ࢖͑ΔpythonϥΠϒϥϦ • Իڹಛ௃நग़ • librosa Իڹಛ௃நग़ (https://librosa.org/) • madmom Իָಛ௃நग़

    (https://madmom.readthedocs.io/en/latest/) • essentia Իָಛ௃நग़ (https://essentia.upf.edu/essentia_python_tutorial.html) • opensmile Իڹಛ௃நग़ (https://www.audeering.com/2020/10/26/introducing- opensmile-3-0/) • ָේσʔλϋϯυϦϯά • Music21 MusicXMLϥΠϒϥϦ (http://web.mit.edu/music21/) • MIDIܥϥΠϒϥϦʢmido, pretty-midi, pypianorollʣ • muspy UC San Diegoൃ Իָࣗಈੜ੒πʔϧΩοτ(https://salu133445.github.io/muspy/) • σʔληοτAPI • mirdata ԻָσʔληοτΛಡΈࠐΉAPIϥΠϒϥϦ (https://mirdata.readthedocs.io/en/ stable/)
  32. ͓·͚ᶄɿԻָ৘ใॲཧͰ࢖͑ΔpythonϥΠϒϥϦ • GPUରԠͷDNN༻લॲཧϥΠϒϥϦ • nnAudio: pytorch༻ • kapre: Tensor fl

    ow༻
  33. ͓·͚ᶄɿԻָ৘ใॲཧͰ࢖͑ΔpythonϥΠϒϥϦ • ͦͷଞ • mireval Իָ৘ใॲཧλεΫͷධՁࢦඪͷࢉग़ ʢhttps:// craffel.github.io/mir_eval/ʣ • kymatio

    ΢ΣʔϒϨοτࢄཚͷܭࢉ (kymat.io) • Omnizart ୆࿷ͷAcademia Sinicaൃࣗಈ࠾ේϥΠϒϥϦ (https://pypi.org/project/omnizart/) • FMP-notebook ਆνϡʔτϦΞϧ (https://www.audiolabs- erlangen.de/resources/MIR/FMP/C0/C0.html)
  34. ͓·͚ᶅɿGTTM • GTTMʢGenerative Theory of Tonal Musicʣ • ָۂ͔Β֊૚తͳߏ଄ΛಘΔͨΊͷཧ࿦ •

    ೝ஌తԻָཧ࿦Ͱ͸҉ҙ࣮ݱϞσϧΑΓ΋ϝδϟʔ ʢͳҹ৅ʣ
  35. ͓·͚ᶅɿGTTMΛߏ੒͢Δαϒཧ࿦ • ָۂͷߏ଄ΛಘΔͨΊͷ4ͭͷαϒཧ࿦Ͱߏ੒ • ߏ଄ͷ෼ੳ • 1. άϧʔϐϯάɿϑϨʔζΛάϧʔϓԽ • 2.

    ഥઅߏ଄ɿϏʔτΛͱΔ্Ͱͷॏཁ౓Λ෇༩ • ؆໿ • 3. λΠϜεύϯ؆໿ɿॏཁ౓ʹج͍ͮͯάϧʔϓ಺Ͱ ໦ߏ଄Λ֊૚తʹ࡞͍ͬͯ͘ • 4. Ԇ௕త؆໿ɿ࿨੠ͷஎ؇ɾۓுʹج͍ͮͯɼɹɹɹ άϧʔϓΛ௒͑ͨࢬͷ઀ଓΛߦ͏
  36. ͓·͚ᶅɿGTTMʹΑΔ֊૚ߏ଄ͷ֫ಘ ϞʔπΝϧτ ϐΞϊιφλ K331. ഥઅߏ଄ άϧʔϐϯά ʢଟٛੑ͋Γʣ

  37. ͓·͚ᶅɿGTTMʹΑΔ؆໿ GTTM - RikenAIP ΑΓ ൈਮ https://aip.riken.jp/ uploads/ 775777483bfe107f16 377730a69f330c.pdf