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

[ASJ_23rd_summer_seminar] 高品質ニューラル音声合成×ト×音声マルチスポット再生 -やはり実装 is all we need-

Takuma OKAMOTO
September 01, 2022

[ASJ_23rd_summer_seminar] 高品質ニューラル音声合成×ト×音声マルチスポット再生 -やはり実装 is all we need-

Takuma OKAMOTO

September 01, 2022
Tweet

More Decks by Takuma OKAMOTO

Other Decks in Research

Transcript

  1. ࣗݾ঺հ ߴ඼࣭χϡʔϥϧςΩετԻ੠߹੒ χϡʔϥϧԻ੠೾ܗੜ੒ɾςΩετԻ੠߹੒Ϟσϧͷൺֱ &OEUPFOEςΩετԻ੠߹੒Ϟσϧɿ+&54 σϞ Ի੠Ϛϧνεϙοτ࠶ੜ σϞ ӳޠχϡʔϥϧԻ੠߹੒ʹΑΔٕज़঺հϏσΦ ·ͱΊɾँࣙ એ఻ɿདྷ݄ͷݚڀൃදձͰͷൃද

    ຊ೔ͷൃද  <latexit sha1_base64="L4PyrqHxk2CTu7gOmrBQv3bHSdA=">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</latexit> Acoustic features <latexit sha1_base64="7VC2AVRUd57DasxvKZPhpBtjT3g=">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</latexit> Synthesized speech waveform <latexit sha1_base64="8hREcmWVayksj+PaC6a157dCp+I=">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</latexit> Charactor <latexit sha1_base64="qUnRaiR6m3iaDFJ654EahFC4AVE=">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</latexit> Phoneme <latexit sha1_base64="AfwYcZl3wmBLpimYFgqoK75ilL8=">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</latexit> Frame <latexit sha1_base64="E1XeJ0BBCJ89zTbLTqjjuN90iKc=">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</latexit> Sample <latexit sha1_base64="jno7dgs/QUu0k9LeFnLbon0r+xo=">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</latexit> Uniform <latexit sha1_base64="rDGYKfH6C+YLKCeSJPi63BaqKi0=">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</latexit> upsampling <latexit sha1_base64="qlWhzUN/YdeFWmBxGfD4ezREfOI=">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</latexit> Input text <latexit sha1_base64="FYn7Awh+Az7dpBlqPxKUbr6gXCo=">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</latexit> Phoneme sequence <latexit sha1_base64="7VC2AVRUd57DasxvKZPhpBtjT3g=">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</latexit> Synthesized speech waveform <latexit sha1_base64="IqSt/ifNR6yIZpewGIdriCSotgY=">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</latexit> Intermediate features <latexit sha1_base64="DO/tGzhM6FDHwzN+3GrekbsF/MA=">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</latexit> Encoder <latexit sha1_base64="TWIjWd2UCKcHB+FaTiAm09jDwTg=">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</latexit> Decoder <latexit sha1_base64="ZOiQO7IsgWClmIsJsscIGbCa0lY=">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</latexit> Text-analyzer (G2P) <latexit sha1_base64="XXfvPpYtiWVMDlnqKOUov6Pq470=">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</latexit> (a) Pipeline TTS models <latexit sha1_base64="8eo0iDuaz2ndkam2nhiWDXNZqvY=">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</latexit> (b) Entire end-to-end TTS models <latexit sha1_base64="nqk1NsXM9/2WKZL04Tz9FSF7UeU=">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</latexit> Acoustic models <latexit sha1_base64="jaCPeYFxdCJbFfQTOxhZQR7TLkc=">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</latexit> (including duration models) <latexit sha1_base64="Q/gGfmMV9x8/OnAwRSZUYWRqEps=">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</latexit> Speech waveform <latexit sha1_base64="eesvGJXjXGSDpY6ubenhxPBVVdo=">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</latexit> (=Neural vocoders) <latexit sha1_base64="sTk7BsxY47KtkV7ZhqwVzhpoq6k=">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</latexit> generative models <latexit sha1_base64="faYIMrsw7EzUXPzctHGSgPhkWds=">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</latexit> Non-uniform <latexit sha1_base64="rDGYKfH6C+YLKCeSJPi63BaqKi0=">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</latexit> upsampling <latexit sha1_base64="VFv6qsOHb8lZibHKA0Krf1uyEUo=">AAACb3ichVHLSgMxFD0d3/VVdaEgSLFU2k1JRVQEoeLGZX1UhbaUmTGtwXkxkxa0+AN+gC5c+AAR8TPc+AMu/ARxJQpuXHg7HRAt6g1JTk7uuTlJNMcQnmTsMaS0tXd0dnX3hHv7+gcGI0PDm55ddXWe023Ddrc11eOGsHhOCmnwbcflqqkZfEvbW27sb9W46wnb2pD7Di+aasUSZaGrkqhCYnHJEBXL5JZMliIxlmJ+RFtBOgAxBJG1I9coYAc2dFRhgsOCJGxAhUctjzQYHOKKqBPnEhL+PschwqStUhanDJXYPRortMoHrEXrRk3PV+t0ikHdJWUUcfbAbtgru2e37Il9/Fqr7tdoeNmnWWtquVMaPBpbf/9XZdIssful+tOzRBnzvldB3h2fadxCb+prByev6wtr8foUu2TP5P+CPbI7uoFVe9OvVvnaKcL0Aemfz90KNqdT6dnUzOp0LDMTfEU3xjGJBL33HDJYQRY5OtfBMc5wHnpRRpUJJdpMVUKBZgTfQkl+Ag/tjhQ=</latexit> (=Alignment) <latexit sha1_base64="TFmHRpLvbUYgMV9rWIhjvdkz3lw=">AAACfXichVHLSgMxFD0d3/VVdSO4KZaKCympiIorwY1L7UMFlTIzRg3NPJjJFLToB/gDLlwpiojoT7jxB1z0E8SlghtB70wHREW9IcnJyT03J4nhSuErxhoJraW1rb2jsyvZ3dPb158aGFzxncAzedl0pOOtGbrPpbB5WQkl+Zrrcd0yJF81qgvh/mqNe75w7JLac/mmpe/YYluYuiKqkhoucct1PF2mPe47MgjZdGALVUllWI5Fkf4J8jHIII4lJ3WJDWzBgYkAFjhsKMISOnxq68iDwSVuE3XiPEIi2uc4QJK0AWVxytCJrdK4Q6v1mLVpHdb0I7VJp0jqHinTyLIHdsWe2T27Zo/s7dda9ahG6GWPZqOp5W6l/2i4+PqvyqJZYfdT9adnhW3MRl4FeXcjJryF2dTX9o+fi3OFbH2MnbEn8n/KGuyObmDXXsyLZV44QZI+IP/9uX+Clclcfjo3tTyZmZ+Kv6ITIxjFOL33DOaxiCWU6dxDnOMGt4l3LatNaLlmqpaINUP4EtrMB4iilCg=</latexit> Temporal resolution unit
  2. ࣗݾ঺հ  Ԭຊ୓ຏ ݚڀςʔϚ Իڹ৴߸ॲཧɿಛʹԻ৔ऩ࿥ɾ੍ޚ ϚΠΫϩϗϯɾεϐʔΧΞϨΠ৴߸ॲཧ 
 ೥݄ʙ೥݄ɿ౦๺େֶઌ୺Ի৘ใγεςϜ म࢜ɾത࢜ɾϙευΫ 


    ೥݄ʙ೥݄ɿ/*$5௒ྟ৔ײϓϩδΣΫτˏ/*$5 
 ೥ʙݱࡏɿࣗ਎ͷՊݚඅʴЋ Ϛϧνεϙοτ࠶ੜ !/*$5 Ի੠ॲཧ 
 ೥݄ʙ೥݄ɿԻ੠ର࿩ɾݴޠࣝผ!/*$5 
 ೥݄ʙݱࡏɿχϡʔϥϧωοτϫʔΫΛ༻͍ͨԻ੠߹੒ɾԻ੠೾ܗੜ੒ˏ/*$5 ԻڹֶձͰ͸ɿ೥݄ʙ೥݄ɿ೔ຊԻڹֶձֶੜɾएखϑΥʔϥϜװࣄձୈظ୅දɼଞ΋Ζ΋Ζ ࡢ೥౓ͷαϚʔηϛφʔ ܥྻม׵ͰͰ͖ΔԻ੠ೝࣝɾԻ੠߹੒ʴ㱣࣮૷JTBMMXFOFFE ࢿྉɿIUUQTJTHEEN-;  )6/5&3Y)6/5&3θϊ͍ͪ͡ΌΜ͔ΒֶͿ࠲ӈͷ໏ ੜ֔ݱ໾ Ұ೔ҰɹΩϩɿ೥ؒ૸ߦڑ཭Ωϩ ೥Ҏ্ܧଓத ɼൃදࢿྉग़དྷͯͳ͔ͬͨͷʹࠓே΋΋ͪΖΜɽɽ ଍ͷΘΒ͡ݚڀੜ׆೥໨ˠ͍ͭʹͭͷ੒Ռ͕༥߹ ௚ۙɿ4MBDLͷ/10ਃ੥ͱ͔ .BUUFSNPTU΁ͷҠߦͱ͔
  3. ໰୊ͷ೉͠͞ ೖྗͱग़ྗͷ௕͕͞શવҧ͏ ʮ)FMMPʯ จࣈɼඵ 
 தؒಛ௃ྔ ໿ϑϨʔϜ 
 Ի੠೾ܗ αϯϓϧ

    
 ˞ϑϨʔϜγϑτྔɼαϯϓϦϯάप೾਺L)[ ಉ͡ςΩετΛಉ͡ਓ͕ಡΜͰ΋ຖճҧ͏Ի੠೾ܗ ೖྗʹରͯ͠਺ઍഒ΋ͷग़ྗΛ֬཰తʹٻΊΔ೉͠͞ ஈ֊ΞοϓαϯϓϦϯάʹΑΔ࣮ݱ ֤จࣈʹର͢Δதؒಛ௃ྔɾԻ੠೾ܗͷ௕͞͸ҟͳΔ ςΩετˠதؒಛ௃ྔɿෆۉҰͳΞοϓαϯϓϦϯά 
 ֤จࣈͷ௕͞Λਪఆ͢Δܧଓ௕ϞσϧΛֶश தؒಛ௃ྔˠԻ੠೾ܗɿۉҰͳΞοϓαϯϓϦϯά ςΩετԻ੠߹੒   <latexit sha1_base64="L4PyrqHxk2CTu7gOmrBQv3bHSdA=">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</latexit> Acoustic features <latexit sha1_base64="7VC2AVRUd57DasxvKZPhpBtjT3g=">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</latexit> Synthesized speech waveform <latexit sha1_base64="8hREcmWVayksj+PaC6a157dCp+I=">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</latexit> Charactor <latexit sha1_base64="qUnRaiR6m3iaDFJ654EahFC4AVE=">AAACanichVG7SgNBFD1ZXzG+YmwUm2BUrMJdERUrwcYyJiYRVGR3HZMh+2J3E9DgD9hZCVopiIifYeMPWPgJop2CjYU3mwVRUe8wM2fO3HPnzIzumtIPiB5iSkdnV3dPvDfR1z8wOJQcTpV8p+4Zomg4puNt6JovTGmLYiADU2y4ntAs3RRlvbbS2i83hOdLx14P9l2xbWkVW+5JQwuYKueqji0ssZPMUJbCSP8EagQyiCLnJK+whV04MFCHBQEbAWMTGnxum1BBcJnbRpM5j5EM9wUOkWBtnbMEZ2jM1nis8GozYm1et2r6odrgU0zuHivTmKJ7uqYXuqMbeqT3X2s1wxotL/s8622tcHeGjkYLb/+qLJ4DVD9Vf3oOsIfF0Ktk727ItG5htPWNg5OXwlJ+qjlNF/TE/s/pgW75Bnbj1bhcE/kzJPgD1O/P/ROUZrPqfHZubTazPBd9RRzjmMAMv/cClrGKHIqhu2Oc4iz2rKSUMWW8narEIs0IvoQy+QGFQYyB</latexit> Phoneme <latexit sha1_base64="AfwYcZl3wmBLpimYFgqoK75ilL8=">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</latexit> Frame <latexit sha1_base64="E1XeJ0BBCJ89zTbLTqjjuN90iKc=">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</latexit> Sample <latexit sha1_base64="jno7dgs/QUu0k9LeFnLbon0r+xo=">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</latexit> Uniform <latexit sha1_base64="rDGYKfH6C+YLKCeSJPi63BaqKi0=">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</latexit> upsampling <latexit sha1_base64="qlWhzUN/YdeFWmBxGfD4ezREfOI=">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</latexit> Input text <latexit sha1_base64="FYn7Awh+Az7dpBlqPxKUbr6gXCo=">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</latexit> Phoneme sequence <latexit sha1_base64="7VC2AVRUd57DasxvKZPhpBtjT3g=">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</latexit> Synthesized speech waveform <latexit sha1_base64="IqSt/ifNR6yIZpewGIdriCSotgY=">AAACenichVHLSgMxFD0d3/XRqgiCm2JRFKGkUlRcCW50p9aqoFJmxtsaOi9m0oIWf8AfcOFKQUT9DDf+gAs/QVxWcOPC2+mAqKg3JDk5uefmJDE8SwZKiKeY1tbe0dnV3RPv7esfSCQHh7YCt+qbVDBdy/V3DD0gSzpUUFJZtOP5pNuGRdtGZbm5v10jP5Cus6mOPNq39bIjS9LUFVPF5Miqo8i36UDqilIl0lXVp6CYTIuMCCP1E2QjkEYUa27yGns4gAsTVdggOFCMLegIuO0iCwGPuX3UmfMZyXCfcII4a6ucRZyhM1vhscyr3Yh1eN2sGYRqk0+xuPusTGFCPIob0RAP4k48i/dfa9XDGk0vRzwbLS15xcTpaP7tX5XNs8Lhp+pPzwolLIReJXv3QqZ5C7Olrx2fNfKLGxP1SXEpXtj/hXgS93wDp/ZqXq3Txjni/AHZ78/9E2zNZrJzmdz6bHopF31FN8Ywjil+73ksYQVrKPC5dVzgFnexd21cm9ZmWqlaLNIM40touQ+AiJLM</latexit> Intermediate features <latexit sha1_base64="DO/tGzhM6FDHwzN+3GrekbsF/MA=">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</latexit> Encoder <latexit sha1_base64="TWIjWd2UCKcHB+FaTiAm09jDwTg=">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</latexit> Decoder <latexit sha1_base64="ZOiQO7IsgWClmIsJsscIGbCa0lY=">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</latexit> Text-analyzer (G2P) <latexit sha1_base64="XXfvPpYtiWVMDlnqKOUov6Pq470=">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</latexit> (a) Pipeline TTS models <latexit sha1_base64="8eo0iDuaz2ndkam2nhiWDXNZqvY=">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</latexit> (b) Entire end-to-end TTS models <latexit sha1_base64="nqk1NsXM9/2WKZL04Tz9FSF7UeU=">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</latexit> Acoustic models <latexit sha1_base64="jaCPeYFxdCJbFfQTOxhZQR7TLkc=">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</latexit> (including duration models) <latexit sha1_base64="Q/gGfmMV9x8/OnAwRSZUYWRqEps=">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</latexit> Speech waveform <latexit sha1_base64="eesvGJXjXGSDpY6ubenhxPBVVdo=">AAACd3ichVHLSgMxFD0d3/U11o3gwmKx1E3JlKIiCIIbV1IfVUGlzExjHZwX88Ja/AF/wIUgKEgVP8ONP+DCTxCXCiK48M50QFTUG5KcnNxzc5Iotq65HmMPCaGtvaOzq7sn2dvXPzAoDqXWXct3VF5WLd1yNhXZ5bpm8rKneTrftB0uG4rON5T9hXB/I+COq1nmmle3+Y4h10xtV1Nlj6iKmMrNLXHfkfV0YKlWlTInK2KG5VkU6Z9AikEGcZQssYltVGFBhQ8DHCY8wjpkuNS2IIHBJm4HDeIcQlq0z3GEJGl9yuKUIRO7T2ONVlsxa9I6rOlGapVO0ak7pExjgt2za/bM7tgNe2Tvv9ZqRDVCL3WalZaW25XB45HV139VBs0e9j5Vf3r2sIuZyKtG3u2ICW+htvTB4cnz6uzKRCPLLtgT+T9nD+yWbmAGL+rlMl85RZI+QPr+3D/BeiEvTeWLy4XMfDH+im6MYhw5eu9pzGMRJZTp3AOcoYmrxJswJmSFXCtVSMSaYXwJQfoAMYKQzg==</latexit> (=Neural vocoders) <latexit sha1_base64="sTk7BsxY47KtkV7ZhqwVzhpoq6k=">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</latexit> generative models <latexit sha1_base64="faYIMrsw7EzUXPzctHGSgPhkWds=">AAACbnichVHLSsNAFD2Nr1ofrQoiiCgWxY1lUoqKq4IbV1IfVaGKJHHUoUkm5FGoxR9wLy4ERUFE/Aw3/oALP0HcCApuXHibBkRFvSGZc8/cc3Pmju6YwvMZe4gpTc0trW3x9kRHZ1d3MtXTu+rJwDV40ZCmdNd1zeOmsHnRF77J1x2Xa5Zu8jW9PFffX6tw1xPSXvGrDt+0tF1b7AhD84kqLUh7MqBcutZWKs0yLIyRn0CNQBpRFGTqChvYhoSBABY4bPiETWjw6ClBBYND3CZqxLmERLjPcYAEaQOq4lShEVum7y5lpYi1Ka/39EK1QX8x6XVJOYIxds+u2Qu7Yzfskb3/2qsW9qh7qdKqN7Tc2UoeDiy//auyaPWx96n607OPHcyEXgV5d0Kmfgqjoa/sH78szy6N1cbZBXsi/+fsgd3SCezKq3G5yJdOkKALUL+P+ydYzWbUqUxuMZvO56KriGMQo5igeU8jj3kUUAwndoRTnMWelX5lSBlulCqxSNOHL6FMfAA/vI41</latexit> Non-uniform <latexit sha1_base64="rDGYKfH6C+YLKCeSJPi63BaqKi0=">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</latexit> upsampling <latexit sha1_base64="VFv6qsOHb8lZibHKA0Krf1uyEUo=">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</latexit> (=Alignment) <latexit sha1_base64="TFmHRpLvbUYgMV9rWIhjvdkz3lw=">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</latexit> Temporal resolution unit
  4. ਂ૚೾ܗੜ੒ϞσϧɿԻڹಛ௃ྔˠԻ੠೾ܗ ࣗݾճؼϞσϧɿࣗΒͷաڈͷ೾ܗσʔλΛ࢖ͬͯαϯϓϧͣͭ೾ܗΛੜ੒ 'MPXܕϞσϧɿന৭ࡶԻΛগͣͭ͠Ի੠೾ܗ΁ͱม׵ ఢରతੜ੒ϞσϧɿࣝผثΛ᱐͢Α͏ʹࣗવͳԻ੠೾ܗΛੜ੒ ֦ࢄ֬཰Ϟσϧɿന৭ࡶԻΛগͣͭ͠ࡶԻআڈ͠ͳ͕ΒԻ੠೾ܗΛੜ੒ ˞࠷ۙྲྀߦΓͷ4UBCMFEJ ff VTJPOܥ 
 


    ೔ຊԻڹֶձࡶղઆهࣄɿIUUQTXXXKTUBHFKTUHPKQBSUJDMFKBTK@@BSUJDMFDIBSKB ೥݄ΑΓΦʔϓϯΞΫηε ੜ੒αϯϓϧσϞϖʔδɿIUUQTXXXPLBNPUPDBNFSBDPNOFVSBM@TQFFDI@TZOUIFTJTJOEFYIUNM χϡʔϥϧԻ੠೾ܗੜ੒Ϟσϧͷൺֱ   <latexit sha1_base64="IGMChlbsVRi5JP0W+/rnhKI5Hfk=">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</latexit> Acoustic features h <latexit sha1_base64="uGbHMILs1w/uDVfLuSQdAwKuipg=">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</latexit> or <latexit sha1_base64="TewyobNoikPRCUbDC4vQ570JLbM=">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</latexit> RNN layers <latexit sha1_base64="S0Aq7JSRliOPJApqWBhAw/wzv0Q=">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</latexit> Sampling <latexit sha1_base64="XwAJggLaXX93Lia51ShnL+x3gd8=">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</latexit> Dilated CNN layers <latexit sha1_base64="9phZoiwXkW+ML/4oj1yXv8aj+Qo=">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</latexit> (Causal) <latexit sha1_base64="bLwU6gFJt97cdAnkxDkEK5r/iAs=">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</latexit> Previous samples <latexit sha1_base64="FmLj/GPf+weBxnRI3XYM1v0ENdg=">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</latexit> Upsampling <latexit sha1_base64="QCXyBGZqqkkxs7xF7ydYqdfH1/w=">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</latexit> Conditioning <latexit sha1_base64="TIsIuRNR7iIACQhl/T0I1MrtuhI=">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</latexit> . . . <latexit sha1_base64="AkEXXrioSrjWQ1Te4BSGJhy7EFM=">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</latexit> In inferece <latexit sha1_base64="yUOIt7Dn3fnN0a1Jq84ZMB+fyz4=">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</latexit> e.g. WaveNet, WaveRNN, LPCNet <latexit sha1_base64="p1zzV1dir6yyIbxLANUjw0qjats=">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</latexit> Probability p(xt |x0, · · · , xt 1, h) <latexit sha1_base64="65JBAGqRA5fPP0gZBUJDbCrQ0rk=">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</latexit> Next sample xt <latexit sha1_base64="folKXAjDyDeKOWUrhrJHMQGOLE8=">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</latexit> (a) Auto-regressive models <latexit sha1_base64="RQlUWz7OdEkafrXAPmZRXxiNUtE=">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</latexit> Gaussian noise <latexit sha1_base64="FmLj/GPf+weBxnRI3XYM1v0ENdg=">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</latexit> Upsampling <latexit sha1_base64="QCXyBGZqqkkxs7xF7ydYqdfH1/w=">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</latexit> Conditioning <latexit sha1_base64="TIsIuRNR7iIACQhl/T0I1MrtuhI=">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</latexit> . . . <latexit sha1_base64="XwAJggLaXX93Lia51ShnL+x3gd8=">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</latexit> Dilated CNN layers <latexit sha1_base64="rfESrVMKxLogTbwPtad+lZv5Tyo=">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</latexit> (Non-causal) <latexit sha1_base64="RQlUWz7OdEkafrXAPmZRXxiNUtE=">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</latexit> Gaussian noise <latexit sha1_base64="DKsKKnk/cqLYqd4QQcjAqsmhdcI=">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</latexit> Invertible layers <latexit sha1_base64="FmLj/GPf+weBxnRI3XYM1v0ENdg=">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</latexit> Upsampling <latexit sha1_base64="QCXyBGZqqkkxs7xF7ydYqdfH1/w=">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</latexit> Conditioning <latexit sha1_base64="TIsIuRNR7iIACQhl/T0I1MrtuhI=">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</latexit> . . . <latexit sha1_base64="FmLj/GPf+weBxnRI3XYM1v0ENdg=">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</latexit> Upsampling <latexit sha1_base64="LYGStWP+lxhs0TBF+daCK7D9eLw=">AAACb3ichVHLSgMxFD0dX7U+WnWhIMhgURShpKWouCp046rUR1VQkZkxraHzYmZaqMUf8AN04cIHiIif4cYfcOEniCtRcOPCO9MBUVFvSHJycs/NSaLaunA9xh4iUlt7R2dXtDvW09vXH08MDK65Vs3ReEmzdMvZUBWX68LkJU94Ot+wHa4Yqs7X1Wre31+vc8cVlrnqNWy+bSgVU5SFpnhEbc3I+UJB1pUGpewkkizFgpB/gnQIkgijaCWusIVdWNBQgwEOEx5hHQpcaptIg8EmbhtN4hxCItjnOECMtDXK4pShEFulsUKrzZA1ae3XdAO1Rqfo1B1Syphg9+yavbA7dsMe2fuvtZpBDd9Lg2a1peX2TvxwZOXtX5VBs4e9T9Wfnj2UMR94FeTdDhj/FlpLX98/fllZWJ5oTrIL9kT+z9kDu6UbmPVX7XKJL58gRh+Q/v7cP8FaJpWeTWWXMslcNvyKKEYxjil67znksIgiSnSujSOc4izyLA1LY5LcSpUioWYIX0Ka/gBlho3B</latexit> + CNN layers <latexit sha1_base64="FmLj/GPf+weBxnRI3XYM1v0ENdg=">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</latexit> Upsampling <latexit sha1_base64="LYGStWP+lxhs0TBF+daCK7D9eLw=">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</latexit> + CNN layers <latexit sha1_base64="TIsIuRNR7iIACQhl/T0I1MrtuhI=">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</latexit> . . . <latexit sha1_base64="TIsIuRNR7iIACQhl/T0I1MrtuhI=">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</latexit> . . . <latexit sha1_base64="vfmLOLXtqta1QaYSGCJQOC8+BGs=">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</latexit> Di↵usion probabilistic models <latexit sha1_base64="QCXyBGZqqkkxs7xF7ydYqdfH1/w=">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</latexit> Conditioning <latexit sha1_base64="StAXOkWPVpOsJ/cWmuo+LufTzU4=">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</latexit> Original waveform x0 <latexit sha1_base64="jOpeWv1fJPFpINsVBFLELFnKmbg=">AAACg3ichVHLSgMxFD2O7/qquhEEKRZFEEpaioogFFzo0lcf0IrMjFFDpzPDJC1q6c6VP+DClYKIutQ/cOMPuPATxKWCGxfeGQdERb0hycnJPTcnieFaQirGHpq05pbWtvaOzkhXd09vX7R/ICedqmfyrOlYjlcwdMktYfOsEsriBdfjesWweN4oz/v7+Rr3pHDsNbXn8vWKvm2LLWHqiqiN6EhJ8V1VX9CrUgrdjtmOkDzWKHFXCstPiLMECyL2EyRDEEcYS070HCVswoGJKirgsKEIW9AhqRWRBINL3DrqxHmERLDP0UCEtFXK4pShE1umcZtWxZC1ae3XlIHapFMs6h4pYxhj9+yCPbM7dsUe2duvtepBDd/LHs3Gh5a7G32HQ6uv/6oqNCvsfKr+9KywhZnAqyDvbsD4tzA/9LX9o+fV2ZWx+jg7ZU/k/4Q9sFu6gV17Mc+W+coxIvQBye/P/RPkUonkVCK9nI5n0uFXdGAYo5ig955GBotYQpbOPcAlrnGjtWqTWkoLc7WmUDOIL6HNvQNmipXX</latexit> Gaussian noise ✏ <latexit sha1_base64="ttQdcKvJB4fb7HaNj5pcWz6/Alw=">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</latexit> p ¯ ↵x0 + p 1 ¯ ↵✏ <latexit sha1_base64="r8UPnsNQGQDK6N68HRr8Hhg4hcM=">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</latexit> Noise level p ¯ ↵ <latexit sha1_base64="jOpeWv1fJPFpINsVBFLELFnKmbg=">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</latexit> Gaussian noise ✏ <latexit sha1_base64="G/iBDbwY2Q1QmnRkXQk/Zob+K0Q=">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</latexit> In training <latexit sha1_base64="017IwdqqqXL28WzJngmgsiDT9Cg=">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</latexit> In inference <latexit sha1_base64="G/iBDbwY2Q1QmnRkXQk/Zob+K0Q=">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</latexit> In training <latexit sha1_base64="5UsE63s3h3d9YRTcaxdiFcnde90=">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</latexit> e.g. ClariNet, Parallel WaveGAN <latexit sha1_base64="lVssHpkIwOyCDqPTwSdIf720VaU=">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</latexit> e.g. WaveGlow, WaveFlow <latexit sha1_base64="q7eIPX8SJzJMAhbcTtylVZ0Dg84=">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</latexit> e.g. MelGAN, HiFi-GAN <latexit sha1_base64="0Glq1FoS1JOYUWmS4q+jUHTTz+c=">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</latexit> e.g. WaveGrad, Di↵Wave, BDDM <latexit sha1_base64="EkH00hyA9UygSQ2SAy6Ooi6rgEc=">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</latexit> (b) CNN-based models <latexit sha1_base64="XEzDLSGsQa47qjmHVvDn11ybH5A=">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</latexit> (c) Flow-based models <latexit sha1_base64="30mXFofZGhecgGeQg3yTOT8gMs4=">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</latexit> (d) Upsampling-based models <latexit sha1_base64="vA66CalmFojIeBLGTIC/dCRn918=">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</latexit> (e) Di↵usion probabilistic models <latexit sha1_base64="TP4cOnnWrJKngxzwawWHKKL/Kig=">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</latexit> Parallel waveform generative models <latexit sha1_base64="Q/gGfmMV9x8/OnAwRSZUYWRqEps=">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</latexit> Speech waveform <latexit sha1_base64="Q/gGfmMV9x8/OnAwRSZUYWRqEps=">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</latexit> Speech waveform <latexit sha1_base64="Q/gGfmMV9x8/OnAwRSZUYWRqEps=">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</latexit> Speech waveform
  5. B ύΠϓϥΠϯϞσϧ ԻڹϞσϧ ςΩετˠԻڹಛ௃ྔ ͱ೾ܗੜ੒Ϟσϧ Իڹ 
 ಛ௃ྔˠԻ੠೾ܗ ΛͦΕͧΕผʑʹֶश 


    ɾԻڹϞσϧɿ5BDPUSPOɼ'BTU4QFFDIʜ 
 ɾ೾ܗੜ੒Ϟσϧɿ8BWF3//ɼ8BWF(MPXɼ)J'J("/ʜ தؒಛ௃ྔ͸ϝϧεϖΫτϩάϥϜ͕࢖ΘΕΔ͜ͱ͕ଟ͍ 
 தؒಛ௃ྔͷਪఆޡࠩʹΑΔԻ࣭ྼԽ 
 ͦ΋ͦ΋࠷దͳதؒಛ௃ྔͳͷ͔ 
 ɹҐ૬৘ใ͍Βͳ͍ ϝϧϑΟϧλόϯΫͰ͍͍ͷ  C &OEUPFOEϞσϧ ςΩετˠԻ੠೾ܗΛͭͷχϡʔϥϧωοτͰ࣮ݱ 
 ɾ'BTU4QFFDI ɼ8BWF5BDPUSPOɼ7*54ɼ+&54ʜ தؒಛ௃ྔ͸࠷ऴతͳԻ੠೾ܗΛग़ྗ͢ΔͨΊʹ࠷దͳ 
 දݱΛχϡʔϥϧωοτ͕ࣗಈతʹ֫ಘ 
 ύΠϓϥΠϯϞσϧΑΓ΋ߴԻ࣭ χϡʔϥϧςΩετԻ੠߹੒Ϟσϧͷൺֱ   <latexit sha1_base64="L4PyrqHxk2CTu7gOmrBQv3bHSdA=">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</latexit> Acoustic features <latexit sha1_base64="7VC2AVRUd57DasxvKZPhpBtjT3g=">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</latexit> Synthesized speech waveform <latexit sha1_base64="8hREcmWVayksj+PaC6a157dCp+I=">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</latexit> Charactor <latexit sha1_base64="qUnRaiR6m3iaDFJ654EahFC4AVE=">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</latexit> Phoneme <latexit sha1_base64="AfwYcZl3wmBLpimYFgqoK75ilL8=">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</latexit> Frame <latexit sha1_base64="E1XeJ0BBCJ89zTbLTqjjuN90iKc=">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</latexit> Sample <latexit sha1_base64="jno7dgs/QUu0k9LeFnLbon0r+xo=">AAACanichVHLSsNAFD2Nr1ofrXWjuBFrxVWZlKLiquDGpa/YQi0lidMamhdJWqjFH3DnSrArBRHxM9z4Ay78BNFdBTcuvEkDokW9w8ycOXPPnTMziq1rrsfYU0QYGBwaHomOxsbGJybjiankvms1HJVLqqVbTlGRXa5rJpc8zdN50Xa4bCg6Lyj1DX+/0OSOq1nmnteyedmQa6ZW1VTZI6ogEbYco5JIsQwLYr4fiCFIIYwtK3GDAxzCgooGDHCY8AjrkOFSK0EEg01cGW3iHEJasM9xghhpG5TFKUMmtk5jjValkDVp7dd0A7VKp+jUHVLOI80e2S3rsgd2x57Zx6+12kEN30uLZqWn5XYlfjqz+/6vyqDZw9GX6k/PHqpYC7xq5N0OGP8Wak/fPD7v7q7vpNtL7Iq9kP9L9sTu6QZm80293uY7HcToA8Sfz90P9rMZcSWT286m8rnwK6KYwwKW6b1XkccmtiAF7s5wgU7kVUgKs8JcL1WIhJppfAth8ROtZoyV</latexit> Uniform <latexit sha1_base64="rDGYKfH6C+YLKCeSJPi63BaqKi0=">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</latexit> upsampling <latexit sha1_base64="qlWhzUN/YdeFWmBxGfD4ezREfOI=">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</latexit> Input text <latexit sha1_base64="FYn7Awh+Az7dpBlqPxKUbr6gXCo=">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</latexit> Phoneme sequence <latexit sha1_base64="7VC2AVRUd57DasxvKZPhpBtjT3g=">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</latexit> Synthesized speech waveform <latexit sha1_base64="IqSt/ifNR6yIZpewGIdriCSotgY=">AAACenichVHLSgMxFD0d3/XRqgiCm2JRFKGkUlRcCW50p9aqoFJmxtsaOi9m0oIWf8AfcOFKQUT9DDf+gAs/QVxWcOPC2+mAqKg3JDk5uefmJDE8SwZKiKeY1tbe0dnV3RPv7esfSCQHh7YCt+qbVDBdy/V3DD0gSzpUUFJZtOP5pNuGRdtGZbm5v10jP5Cus6mOPNq39bIjS9LUFVPF5Miqo8i36UDqilIl0lXVp6CYTIuMCCP1E2QjkEYUa27yGns4gAsTVdggOFCMLegIuO0iCwGPuX3UmfMZyXCfcII4a6ucRZyhM1vhscyr3Yh1eN2sGYRqk0+xuPusTGFCPIob0RAP4k48i/dfa9XDGk0vRzwbLS15xcTpaP7tX5XNs8Lhp+pPzwolLIReJXv3QqZ5C7Olrx2fNfKLGxP1SXEpXtj/hXgS93wDp/ZqXq3Txjni/AHZ78/9E2zNZrJzmdz6bHopF31FN8Ywjil+73ksYQVrKPC5dVzgFnexd21cm9ZmWqlaLNIM40touQ+AiJLM</latexit> Intermediate features <latexit sha1_base64="DO/tGzhM6FDHwzN+3GrekbsF/MA=">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</latexit> Encoder <latexit sha1_base64="TWIjWd2UCKcHB+FaTiAm09jDwTg=">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</latexit> Decoder <latexit sha1_base64="ZOiQO7IsgWClmIsJsscIGbCa0lY=">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</latexit> Text-analyzer (G2P) <latexit sha1_base64="XXfvPpYtiWVMDlnqKOUov6Pq470=">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</latexit> (a) Pipeline TTS models <latexit sha1_base64="8eo0iDuaz2ndkam2nhiWDXNZqvY=">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</latexit> (b) Entire end-to-end TTS models <latexit sha1_base64="nqk1NsXM9/2WKZL04Tz9FSF7UeU=">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</latexit> Acoustic models <latexit sha1_base64="jaCPeYFxdCJbFfQTOxhZQR7TLkc=">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</latexit> (including duration models) <latexit sha1_base64="Q/gGfmMV9x8/OnAwRSZUYWRqEps=">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</latexit> Speech waveform <latexit sha1_base64="eesvGJXjXGSDpY6ubenhxPBVVdo=">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</latexit> (=Neural vocoders) <latexit sha1_base64="sTk7BsxY47KtkV7ZhqwVzhpoq6k=">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</latexit> generative models <latexit sha1_base64="faYIMrsw7EzUXPzctHGSgPhkWds=">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</latexit> Non-uniform <latexit sha1_base64="rDGYKfH6C+YLKCeSJPi63BaqKi0=">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</latexit> upsampling <latexit sha1_base64="VFv6qsOHb8lZibHKA0Krf1uyEUo=">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</latexit> (=Alignment) <latexit sha1_base64="TFmHRpLvbUYgMV9rWIhjvdkz3lw=">AAACfXichVHLSgMxFD0d3/VVdSO4KZaKCympiIorwY1L7UMFlTIzRg3NPJjJFLToB/gDLlwpiojoT7jxB1z0E8SlghtB70wHREW9IcnJyT03J4nhSuErxhoJraW1rb2jsyvZ3dPb158aGFzxncAzedl0pOOtGbrPpbB5WQkl+Zrrcd0yJF81qgvh/mqNe75w7JLac/mmpe/YYluYuiKqkhoucct1PF2mPe47MgjZdGALVUllWI5Fkf4J8jHIII4lJ3WJDWzBgYkAFjhsKMISOnxq68iDwSVuE3XiPEIi2uc4QJK0AWVxytCJrdK4Q6v1mLVpHdb0I7VJp0jqHinTyLIHdsWe2T27Zo/s7dda9ahG6GWPZqOp5W6l/2i4+PqvyqJZYfdT9adnhW3MRl4FeXcjJryF2dTX9o+fi3OFbH2MnbEn8n/KGuyObmDXXsyLZV44QZI+IP/9uX+Clclcfjo3tTyZmZ+Kv6ITIxjFOL33DOaxiCWU6dxDnOMGt4l3LatNaLlmqpaINUP4EtrMB4iilCg=</latexit> Temporal resolution unit
  6. +&54 IUUQTBSYJWPSHBCT  ԻڹϞσϧ 'BTU4QFFDI ͱ೾ܗੜ੒Ϟσϧ )J'J("/ Λ 
 ϝϧεϖΫτϩάϥϜΛհͣ͞ʹܨ͛ͯಉ࣌ʹֶश

    ैདྷͷԻ੠߹੒ɼػց຋༁ɼը૾ੜ੒ͰഓΘΕٕͨज़ͷूେ੒ 5SBOTGPSNFS ࣗݾ஫ҙػߏ ɿػց຋༁ ԻૉΞϥΠϝϯτ 'PSXBSECBDLXBSEΞϧΰϦζϜ 
 ɹɹɹɹɹɹɹɹɹɹɿӅΕϚϧίϑܕԻ੠ೝࣝԻ੠߹੒ ఢରతੜ੒Ϟσϧ )J'J("/ ɿը૾ੜ੒ &41OFU554ʹ͓͚Δ࣮૷͋Γ 
 IUUQTHJUIVCDPNFTQOFUFTQOFUUSFFNBTUFSFTQOFUHBO@UUTKFUT 
 FTQOFUFHTMKTQFFDIUUTDPOGUVOJOHUSBJO@KFUTZBNM ࣌ؒҎ಺ͷσʔλ $.6"3$5*$ ΍αϯϓϦϯάप೾਺ 
 L)[Ͱ΋ಈ͘ެ։ίʔύεͷΈͰ࣮૷Մೳ 
 IUUQTXXXPLBNPUPDBNFSBDPNOFVSBM@TQFFDI@TZOUIFTJTJOEFYIUNM 
 
 ˞ͦͷ্Ͱɼެࣜൃද͸·ͩ ݄ͷ*OUFSTQFFDI &OEUPFOEχϡʔϥϧԻ੠߹੒Ϟσϧͷྫ   <latexit sha1_base64="qlWhzUN/YdeFWmBxGfD4ezREfOI=">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</latexit> Input text <latexit sha1_base64="FYn7Awh+Az7dpBlqPxKUbr6gXCo=">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</latexit> Phoneme sequence <latexit sha1_base64="eqMndbZ87wqsy7S2OybjDDfQGcs=">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</latexit> Output raw speech waveform <latexit sha1_base64="//pmT2HVncuEMTe5lL0FgNladLA=">AAACdHichVHLSsNAFD2Nr1ofrboRdFEMFTeWqYiKq4IbN4KtthW0lCSONTQvkmmhFn/AH3DRlaKI+Blu/AEXfoK4rOjGhTdpQLSoN0zmzJl77pyZqzqG7gnGniJSX//A4FB0ODYyOjYeT0xMFj277mq8oNmG7e6piscN3eIFoQuD7zkuV0zV4CW1tuHvlxrc9XTb2hVNh5dNpWrpR7qmCKIqifgWNxY9h2vCtauuYlYSMkuzIJK9IBMCGWFs24kbHOAQNjTUYYLDgiBsQIFH3z4yYHCIK6NFnEtID/Y5ThEjbZ2yOGUoxNboX6XVfshatPZreoFao1MMGi4pk0ixR3bLOuyB3bFn9vFrrVZQw/fSpFntarlTiZ9N77z/qzJpFjj+Uv3pWeAIa4FXnbw7AePfQuvqGyfnnZ31fKo1zy7ZC/m/YE/snm5gNV616xzPtxGjBmR+PncvKC6lMyvp5dySnF0OWxHFDOawQO+9iiw2sY1C0JM2rnAdeZNmJVlKdVOlSKiZwreQ0p+jrpAo</latexit> Mel-spectrogram <latexit sha1_base64="BQ7TlRDT2S3gLQotOTEfP+orRfo=">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</latexit> Text analyzer <latexit sha1_base64="7/xh7p6SaIPVWfg3H1nsgMjD+lE=">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</latexit> Transformer-based decoder <latexit sha1_base64="PnyUOmBroUC9Arfpf/xMC1qClTI=">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</latexit> Embedding layer <latexit sha1_base64="4MdmVFiF6d7lamJ930mNAt0130I=">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</latexit> Transformer-based encoder <latexit sha1_base64="BT+hlArdaMVLxMAlmUOnhqOUqbc=">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</latexit> Variance adaptor <latexit sha1_base64="Nv77tzbKgPFfE0bu0ljrf+8nTrQ=">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</latexit> FastSpeech2-based acoustic model <latexit sha1_base64="7PgzVEb/3zT9PeaXY3w6A30Z5P4=">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</latexit> speech waveform generator <latexit sha1_base64="Spt1qHBhbJUIkNlfsGfs3+CK8z8=">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</latexit> Alignment <latexit sha1_base64="tWhmS3euCF9hjm94NmDr6V7WeKM=">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</latexit> module <latexit sha1_base64="uPOoCMeI/QGvOmCE/Kgqty0nuO0=">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</latexit> h <latexit sha1_base64="fcv157TIFnfAZ8Qn+HD8KQYt5Uo=">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</latexit> d <latexit sha1_base64="XJ9SuAZSJRKmW5MBOgNbUVfNqwA=">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</latexit> : Training only <latexit sha1_base64="DbQSY09dG+YrFEpQSA/TvuES5EY=">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</latexit> HiFi-GAN-based %-JNFUBM *OUFSTQFFDI ࣝผثχϡʔϥϧωοτ ఢରతֶश ("/
  7. ൵ใ   1Z5PSDI͕...BDͷ.14 "QQMFγϦίϯͷ(16 ʹରԠ࢝͠ΊͨΒ͍͠ 
 IUUQT[FOOEFWIJEFUPTIJBSUJDMFT@QZUPSDINNBDCPPLHQV ͷͰΠΩͬͯ1SFWJFX൛ಋೖͯ͠+&54ര଎Ͱਪ࿦Ͱ͖Δͱࢥ͍͖΍ɼɼɼ DVNTVN

    MFOHUI@SFHVMBUPSQZ ͱUPSDIOOVUJMTXFJHIU@OPSN IJ fi HBOQZ ͕·ͩ.14 ະରԠͩͬͨͨΊ$16 35'ɿ ΑΓ΋஗͍ 35'ɿ ɽɽɽ 
 FYQPSU1:503$)@&/"#-&@.14@'"--#"$,Ͱ࣮ߦ  ͱ͍͏Θ͚Ͱࠓޙʹظ଴
  8. εϐʔΧΞϨΠΛ༻͍ͨϚϧνεϙοτ࠶ੜ Իͷฉ͑͜ΔΤϦΞͷԻѹΛɼฉ͑͜ͳ͍ΤϦΞͷԻѹΛͱͨ͠Ի৔੍ޚٕज़ ݴޠԻ੠Ϛϧνεϙοτ࠶ੜγεςϜ ຋༁͞Ε֤ͨݴޠͷԻ੠͕ͦΕͧΕҧͬͨ৔ॴͰฉ͑͜ΔԻ੠຋༁ͷ৽ͨͳ6*ͱ࣮ͯ͠૷ Ի੠Ϛϧνεϙοτ࠶ੜ   y x z

    r0 rref Φ φs P(rref , φ, 0, ω) = 1 P(rref , φ, 0, ω) = 0 Bright zone Circular loudspeaker array Reference circle Dark Zone )FMMP χϡʔϥϧݴޠࣝผ  χϡʔϥϧԻ੠ೝࣝ χϡʔϥϧ ػց຋༁ Ϛϧνεϙοτ ࠶ੜϑΟϧλ ଟݴޠ χϡʔϥϧ Ի੠߹੒ )FMMP ͜Μʹͪ͸ 㟬޷ উ֞ೞࣁਃ B DIԁܗεϐʔΧ C ݴޠσϞγεςϜ ೔ຊޠ ӳޠ தࠃޠ ؖࠃޠ DI εϐʔΧ DN ࠓճͷσϞ 50LBNPUP 8"41""
  9. ֎෦ࢿۚ ຊݚڀͷҰ෦͸+414Պݚඅ+1,͓Αͼ+45 プ ϩ グ ϥϜϚωʔ ジ ϟʔ 1. ͷҭ੒ɾ׆༂ਪਐ

    プ ϩ グ ϥϜ ࣮ ࢪ୅දऀ඗ాܒଠ ͷॿ੒Λड͚ͨ΋ͷ で ͋Δ ֎෦࿈ܞ ԁܗε ピ ʔΧΞϨΠͷࢼ࡞͸๺೔ຊԻڹגࣜձࣾͱͷڞಉݚڀͱͯ͠ߦΘΕେࢁ৻ೋࢯɼ্໺ࠀ࢘ࢯɼԬ෦࢘ࢯɼ ୩݈ଠ࿠ࢯʹײँͷҙΛද͢Δ ಺෦࿈ܞ Ի੠Ϛϧνεϙοτ࠶ੜٕज़ͷࣾձల։͸/*$5Πϊ ベ ʔγϣϯ デザ ΠϯΠχγΞςΟ ブ ڞ૑ デザ Πϯ プ ϩ ジ ΣΫτ ͱͯ͠ߦΘΕɼ プ ϩ ジ ΣΫτϝϯ バ ʔ で ͋Δ඗ాܒଠࢯɼਢ౻ඒ༏رࢯɼ܂ݪສබࢯʹײँͷҙΛද͢Δ ँࣙ 
  10. χϡʔϥϧԻ੠߹੒ ̍ɿˑࢁԼ ਆށେ./*$5ݚमੜ ɼԬຊɼߴౡ ਆށେ ɼୌޱ ਆށେ ɼށా ໊େ ɼՏҪɼz'$)J'J("/ɿ

    શ݁߹૚ܕΞοϓαϯϓϦϯάΛಋೖͨ͠ߴ଎)J'J("/z ɿˑਗ਼ਫ ਆށେ./*$5ݚमੜ ɼԬຊɼߴౡ ਆށେ ɼୌޱ ਆށେ ɼށా ໊େ ɼՏҪɼzجຊप೾਺ ੍ޚՄೳͳϝϧεϖΫτϩάϥϜೖྗܕ)J'J("/ͷॳظݕ౼z ɿˑ೔ా ࿨Վࢁେ./*$5ݚमੜ ɼԬຊɼ੢ଜ ࿨Վࢁେ ɼށా ໊େ ɼՏҪɼzԻ੠ೝࣝίʔύεΛ༻͍ ͨߴ඼࣭ෳ਺࿩ऀԻ੠߹੒ʹ޲͚ͨχϡʔϥϧϘίʔμʹΑΔଳҬ֦ுz ɿদݪ ݩਆށେ/*$5ݩݚमੜ ɼ˓Ԭຊɼߴౡ ਆށେ ɼୌޱ ਆށେ ɼށా ໊େ ɼՏҪɼz)BSNPOJD /FU ɿߴௐ೾ೖྗͱ-BZFSXJTF2VBTJ1FSJPEJD৞ΈࠐΈΛ༻͍ͨجຊप೾਺੍ޚՄೳͳߴ଎χϡʔϥϧϘίʔμz 2ɿদݪ ݩਆށେ/*$5ݩݚमੜ ɼԬຊɼ˕ߴౡ ਆށେ ɼୌޱ ਆށେ ɼށా ໊େ ɼՏҪɼzߏԻো֐ऀ ͷͨΊͷߴ໌ྎ౓Ի੠߹੒ʹ͓͚Δ)J'J("/Λ༻͍ͨ඼࣭վળz Ի੠ݴޠ֫ಘΤʔδΣϯτ +414Պݚඅج൫$෼୲  2ɿˑখদ ౦޻େ. ɼԬຊɼࣰ࡚ ౦޻େ ɼz࣮Ի੠Λ༻͍ͨԻ੠ݴޠ֫ಘΤʔδΣϯτͷධՁz Ϛϧνεϙοτ࠶ੜ +414Պݚඅج൫$୅ද Ћ  rɿ˓Ԭຊɼେࢁ ๺೔ຊԻڹ ɼ্໺ ๺೔ຊԻڹ ɼԬ෦ ๺೔ຊԻڹ ɼzνϟωϧখܕԁܗεϐʔΧΞϨΠ Λ༻͍ͨϚϧνεϙοτ࠶ੜγεςϜͷ࣮૷z ηογϣϯձ৔ʹͯݴޠϚϧνεϙοτ࠶ੜ࣮ػσϞ͋Γ એ఻ɿࠓ݄ͷݚڀൃදձˏ๺ւಓͰͷൃද 
  11. Thank you for your !! Ԭຊ୓ຏ (Takuma OKAMOTO)ˏNICT e-mail: HP:

    https://www.okamotocamera.com Twitter: @okamotocamerea