Comments
Transcript
歌唱合成音声のハスキーボイス化 A method for transforming normal
Vol.2013-MUS-99 No.36 2013/5/12 ሗฎ⌮Ꮫ◊✲ሗ࿌ IPSJ SIG Technical Report ḷၐྜᡂ㡢ኌࡢࣁࢫ࣮࢟࣎ࢫ బࠎᮌ ᫍኴᮁ†1 ⓑᮌ ၿᑦ †1 ḷၐྜᡂ㡢ኌࡣ㸪㡢ኌྜᡂ࣭ࢹࢫࢡࢺࢵࣉ࣑࣮ࣗࢪࢵࢡ㸦DTM㸧ࢯࣇࢺ࢙࢘➼ࢆ⏝࠸࡚㸪ࢥࣥࣆ࣮ࣗࢱୖ࡛⏕ ᡂࡉࢀࡓḷၐ㡢ኌ࡛࠶ࡿ㸬ᮏ◊✲࡛ࡣ㸪ḷၐྜᡂ㡢ኌࡢࣁࢫ࣮࢟࣎ࢫࡢ᪉ἲࢆᥦࡍࡿ㸬ࣁࢫ࣮࢟࣎ࢫࡢᇶ ᮏⓗ࡞ᡭἲࡣ㸪㡢ኌಙྕࡢኌᖏሗ࡛࠶ࡿṧᕪಙྕฎ⌮ࢆຍ࠼ࡿࠕṧᕪኚ࡛ࠖ࠶ࡿ㸬ࡇࡢࠕṧᕪኚࠖࡘ࠸࡚ ᮏ◊✲࡛ࡣ✀㢮ࡢ᪉ἲࢆᥦࡋ࡚࠸ࡿ㸬ࡉࡽࠕṧᕪኚࠖฎ⌮ᚋࡢ㡢㉁ྥୖࢆᅗࡿࡓࡵࢥࣛࢨ࣮ࡢᛂ⏝ࢆヨ ࡳࡿ㸬ࡲࡓṧᕪಙྕࡢᖜሗࡽࠕṧᕪኚࠖฎ⌮ࡢࣦ࢙ࣞࣝࢆỴᐃࡍࡿࡇࡼࡗ࡚㸪ࣁࢫ࣮࢟࣎ࢫࡢ⮬ື⏕ ᡂࢆ⾜࡞࠺㸬ᥦࡋࡓ✀㢮ࡢࠕṧᕪኚࠖἲ㸪ࢥࣛࢨ࣮ᛂ⏝ἲ㸪⮬ື⏕ᡂἲ㸪ࡢ㸲ࡘࡢᡭἲࢆ⏝࠸࡚ྜᡂࡋࡓ 㡢ኌࡢ⫈ྲྀᐇ㦂ࢆ⾜࠸㸪㨩ຊⓗ࡞ࣁࢫ࣮࢟࣎ࢫࡢࡓࡵࡣࡢࡼ࠺࡞ᡭἲࡸᣦᶆ᭷⏝ᛶࡀ࠶ࡿࡢࡘ࠸࡚㸪 ᳨ドࡍࡿ㸬 A method for transforming normal singing voice into husky synthesized singing voice Seitarou Sasaki†1 and Yoshinao Shiraki†1 The singing synthesized voice , is a singing voice using speech synthesis and the desktop music (DTM) software, etc., is generated on a computer. In this paper, we propose a method for transforming normal singing voice into husky synthesized singing voice. The residual signal is a vocal information of the speech signal. "This conversion to the residual signal" is the basic techniques of husky voice conversion. In this study, we propose two methods for this "residual conversion". In order to improve sound quality of the "residual conversion" treatment, we try to apply the equalizer. Further by determining the Revel "residual conversion" process from the amplitude information of the residual signal, we perform the automatic generation of husky voice. For generating synthesized speech using the four methods of the two types of proposed "residual conversion" method, the equalizer application method, an automatic generation method, we do listening tests. We do verification for two things; Which approach or increase the degree of attraction husky voice ? What indicators suitable to measure the degree of glamor husky voice ? 1. ࡣࡌࡵ 㨩ຊࢆᣢࡘኌࡢ୍ࡘࡋ࡚ᣲࡆࡽࢀࡿࡢࡀࣁࢫ࣮࢟࣎ࢫ ḷၐྜᡂ㡢ኌࡣ㸪㡢ኌྜᡂ࣭ࢹࢫࢡࢺࢵࣉ࣑࣮ࣗࢪࢵ ḷኌࡣ␗࡞ࡾ㸪ࠕࡋࡷࡀࢀࠖࡸࠕࡍࢀࠖ࠸ࡗࡓ୍ぢ㸪 ࢡ㸦DTM㸧ࢯࣇࢺ࢙࢘➼ࢆ⏝ࡋ㸪ࢥࣥࣆ࣮ࣗࢱୖ࡛⏕ ⨾ࡋࡉ┦ࡍࡿࡼ࠺࡞ᛶ㉁ࢆᣢࡘ㸬ࡋࡋ㸪ယឋ࡞࠸ ᡂࡉࢀࡓḷၐ㡢ኌ࡛࠶ࡿ㸬௦⾲ⓗ࡞ࡋ࡚㸪ࢡࣜࣉࢺ࣭ࣥ ࡗࡓே㛫ⓗ㨩ຊࢆឤࡌࡉࡏࡿሙྜࡀ࠶ࡾ㸪ᐇ㝿ࣁࢫ࣮࢟ ࣇ࣮ࣗࢳ࣮࣭࣓ࣕࢹࡽⓎࡉࢀࡓࠕึ㡢࣑ࢡࠖࡸ㣩 ࣎ࢫࢆᣢࡘከࡃࡢ᭷ྡ࡞࣮ࢸࢫࢺࡀᏑᅾࡍࡿ㸬ᮏ◊ ࡛࠶ࡿ㸬ࣁࢫ࣮࢟࣎ࢫࡣ㸪࢜࣌ࣛ࡞࡛⪥ࡍࡿ⨾ࡋ࠸ ᒇ㸭Ⳮⵦ㸦࠶ࡵࡸ࣭࠶ࡸࡵ㸧ไసࡼࡿࣇ࣮࢙ࣜ࢘ࠕUTAUࠖ ✲࡛ࡣ㸪≉㸪ࡇࡢࣁࢫ࣮࢟࣎ࢫ↔Ⅼࢆᙜ࡚㸪ḷၐྜ ࡀᣲࡆࡽࢀࡿ[1]㸬㏆ᖺ㸪ࢹࢫࢡࢺࢵࣉ࣑࣮ࣗࢪࢵࢡ㸦DTM㸧 ᡂ㡢ኌࢆࣁࢫ࣮࢟࣎ࢫࡍࡿ᪉ἲࢆᥦࡍࡿ㸬ࡲࡓ㸪ࣁ ࡸ࢚ࣞࢡࢺࣟࢲࣥࢫ࣑࣮ࣗࢪࢵࢡ㸦EDM㸧ࡢᬑཬ࣭ὶ⾜ ࢫ࣮࢟࣎ࢫࢆ㏻ࡋ㸪ே㛫ⓗ㨩ຊࢆᣦᶆ࣭≉ᚩࡍࡿ క࠸㸪ࡇࢀࡽࡢḷၐྜᡂ㡢ኌࡀ⏝ࡉࢀࡿᶵࡀቑ࠼㸪ㄆ ࡶᮏ◊✲ࡢ┠ⓗ࡛࠶ࡿ㸬 ▱ᗘࡶ㧗ࡲࡗ࡚࠸ࡿ㸬ࡑࡋ࡚㸪ࡇࢀࡽࡢ㡢ኌࢆ⫗ኌ㏆࠸ ᮏ◊✲࡛ࡣ㸪ࣁࢫ࣮࢟࣎ࢫࢆḟࡢࡼ࠺ᤊ࠼ࡿ㸬ࣁࢫ ḷኌࡋ࡚⏝ࡋࡓ࠸࠸࠺㟂せࡀ㧗ࡲࡿࡶணࡉࢀࡿ㸬 ࣮࢟࣎ࢫࡣ㸪ࡍࢀࡓኌ㸪ࢲ࣑ኌ࡛࠶ࡾ㸪ኌᖏ ᮏ◊✲࡛ࡣ⫗ኌ㏆࠸ḷၐྜᡂ㡢ኌࡢ⏕ᡂࢆ┠ᶆ⨨ࡃ㸬 ࡽ⏕ࡌࡿ㞧㡢ࢆྵࡴኌ࡛࠶ࡿ㸬ኌᖏࡢჾ㉁ⓗኚࡸ㐠ື㯞 ┠ᶆ㐩ᡂࡢࡓࡵࡢ୍ࡘࡢഃ㠃ࡋ࡚㸪⊂≉࡞ḷኌ㸪ḷ࠸᪉ ⑷ࡽ⏕ࡌࡿ≧࡛࠶ࡿႤኌࡀ㸪ࡍࢀࡓኌ㸪ࢲ࣑ኌ࡞ ࡞ゝࡗࡓᴦ㆕ୖ⾲ࡏ࡞࠸せ⣲㸪࠸ࢃࡤࠕே㛫ⓗ㨩ຊࠖ ࡢ㸪㡢Ⰽ㛵ࡍࡿ⥲ྜⓗ࡞ኌࡢ␗ᖖࢆᣦࡍⅭ[3]㸪ࣁࢫ࣮࢟ ࢆḷၐྜᡂ㡢ኌຍࡍࡿࡀᣲࡆࡽࢀࡿ[2]㸬ࡇࡢே㛫ⓗ ࣎ࢫࡶႤኌࡢ୍ᙧែぢ࡞ࡏࡿ㸬㢮ఝࡋࡓኌࢹࢫ࣎ ࢫ࡞ࡀᣲࡆࡽࢀࡿࡀ㸪ࡇࢀࡣኌᖏࡣูࡢჾᐁࡽ⏕ࡌ †1 ᮾ㑥Ꮫ Toho University [a]ࣁࢫ࣮࢟࣎ࢫࡢ௦⾲ⓗ࡞ࡋ࡚㸪㺙㺙㺼㺻㺓㺎㺷㺛㺞㺎㺛㺼ࡢ᱓⏣ె♸Ặࡢḷኌ ࡢ୍㒊ࢆ⏝ࡋ࡚࠸ࡿ㸬 [b]ࣁࢫ࣮࢟࣎ࢫ࡛ࡣ࡞࠸㡢ኌࡋ࡚➨୍ⴭ⪅⮬㌟ࡢḷኌࢆ⏝ࡋ࡚࠸ࡿ㸬 ⓒ 2013 Information Processing Society of Japan ࡓ㡢ࡀྵࡲࢀࡿኌ࡛࠶ࡿⅭ㸪ࣁࢫ࣮࢟࣎ࢫࡣ༊ูࡉࢀ ࡿࡇࡀከ࠸[4],[5]㸬ࡲࡓᅗ 1 ࡢ♧ࡍࡼ࠺㸪ࣁࢫ࢟ [a] ࣮࣎ࢫࡢ㡢ኌ ࡣࣁࢫ࣮࢟࣎ࢫ࡛ࡣ࡞࠸㡢ኌ[b]ẚ㸪 ࿘Ἴᩘᡂศࡀᗈᖏᇦศᩓศᕸࡋ࡚࠸ࡿࡀ☜ㄆࡉࢀࡿ㸬 1 Vol.2013-MUS-99 No.36 2013/5/12 ሗฎ⌮Ꮫ◊✲ሗ࿌ IPSJ SIG Technical Report ᮏ◊✲࡛ࡣ㸪ࡇࡢ࿘Ἴᩘᡂศࡢᗈᖏᇦศᩓศᕸࢆࣁࢫ࣮࢟ ࡚࠸ࡿ㸬ḟࡢᅗ 2 ࠕṧᕪኚἲࠖࢆᇶᮏࡍࡿࣁࢫ࣮࢟ ࣎ࢫࡢせ࡞≉ᚩぢ࡞ࡍ㸬 ࣎ࢫࡢฎ⌮ࣇ࣮ࣟࢆ♧ࡍ㸬 ࣁࢫ࣮࢟࣎ࢫ ୍⯡ࡢேࡢኌ ᅗ 2 ṧᕪኚἲᇶ࡙ࡃࣁࢫ࣮࢟࣎ࢫࡢฎ⌮ࣇ࣮ࣟ Fig. 2 Processing flow of the husky voice of based on residual conversion method. ᮏ◊✲࡛ࡣ㸪ࡇࡢṧᕪኚἲ㛵ࡋ࡚௨ୗ㏙ࡿࡘࡢ ᡭἲࢆ᳨ウࡍࡿ㸬 ᅗ 1 ࿘Ἴᩘᡂศศᯒࡼࡿࣁࢫ࣮࢟࣎ࢫࡢẚ㍑ᅗ [ c] Fig. 1 Comparison with a normal voice and husky voice with a frequency component analysis. (2) ṧᕪኚἲ㸯㸦ࣀࢬᙉㄪἲ㸧 ኌᖏሗ࡛࠶ࡿṧᕪಙྕࡣ㸪࿘ᮇᛶࡢ࠶ࡿᖜࡢࡁ࠸ ࣃࣝࢫᡂศ㸪ࡑࢀ௨እࡢᖜࡢᑠࡉ࠸ࣀࢬᡂศู ᮏ◊✲࡛ࡣ㸪ḷၐྜᡂ㡢ኌࢆ⥺ᙧண ศᯒ㸦LPC㸧ࡍࡿ ࡛ࡁࡿ㸦ᅗ 3㸧㸬ࣀࢬᙉㄪἲࡣ㸪ࡇࡢࣀࢬᡂศࡢᖜࢆ ࡼࡗ࡚ᚓࡽࢀࡿṧᕪಙྕὀ┠ࡋ㸪ṧᕪಙྕኚ㸪 ቑᖜࡉࡏࡿࡼࡾ㸪ྜᡂᚋࡢ㡢ኌࡢࣀࢬࢆᙉㄪࡉࡏ㸪 ฎ⌮ࢆຍ࠼ࡿࡼࡗ࡚ࣁࢫ࣮࢟࣎ࢫࢆ⾜࠺㸬ࡲࡓ㸪 ࣁࢫ࣮࢟࣎ࢫࢆᅗࡿᡭἲ࡛࠶ࡿ. 㡢㉁㠃㸪ỗ⏝ᛶࢆ⪃៖ࡋࡓ᪉ἲࡢ⪃ࡶ⾜࠺㸬㡢㉁ࡢಖᣢ ࣃࣝࢫᡂศࣀࢬᡂศࡢุᐃࡣ㸪ṧᕪಙྕෆࡢᖜ ࡣࢥࣛࢨ࣮ࡢᢏ⾡ࢆᛂ⏝ࡋ㸪ỗ⏝ᛶࡢ㠃࡛ࡣ㸪ࣁࢫ ࡢ⤯ᑐ್ࢆ⏝ࡍࡿ㸬ᖜࡢ್ࡀ⤯ᑐ್ࡢᖹᆒ௨ୖࡢሙྜ㸪 ࣮࢟࣎ࢫࡢ⮬ື⏕ᡂࢆヨࡳࡿ㸬⮬ື⏕ᡂࡢᢏ⾡ࡣ㸪ᅵᖌ ࣃࣝࢫᡂศุᐃ㸪ᖹᆒ௨ୗࡢሙྜ㸪ࣀࢬᡂศุᐃࡍ ▱⾜,ᴙᩗⶶ,∦ᒸⱥᖾ,⸨ᮌᬸஓⴭࠕṧᕪἼᙧศᯒࡼࡿႤ ࡿ㸦ᅗ 4㸧㸬 ࣃࣝࢫᡂศ ኌࡢᐈほⓗホ౯ࠖᩥ⊩[6]ࢆཧ⪃ࡋ࡚࠸ࡿ㸬ᩥ⊩[6]࡛グ㏙ ࡉࢀࡓࠕႤኌࡢᗘྜ࠸ࢆ ࡿᣦᶆࡋ࡚㸪ᝈ⪅ࡢṧᕪἼᙧ ࡢࣄࢫࢺࢢ࣒ࣛࢆ⏝ࡍࡿࠖ࠸࠺᪉ἲࢆࣄࣥࢺࡋ࡚㸪 ࣁࢫ࣮࢟࣎ࢫ࠾ࡅࡿ㨩ຊᗘࢆィࡿᣦᶆ㸪⮬ື⏕ᡂࡢ ᡭἲࢆ⪃ࡍࡿ㸬᭱ᚋ㸪ᥦࡋࡓ✀㢮ࡢࠕṧᕪኚࠖ ἲ㸪ࢥࣛࢨ࣮ᛂ⏝ἲ㸪⮬ື⏕ᡂἲ㸪ࡢ㸲ࡘࡢᡭἲࢆ⏝ ࠸࡚ྜᡂࡋࡓ㡢ኌ࡛⫈ྲྀᐇ㦂ࢆ⾜࠸㸪ࡑࡢ᭷⏝ᛶ㛵ࡋ࡚ ᐇドࢆヨࡳࡿ㸬 ࣀࢬᡂศ 2. ࣁࢫ࣮࢟࣎ࢫࡢᇶᮏⓗᡭἲ ᅗ 3 ࣃࣝࢫᡂศࣀࢬᡂศࡢ࣓࣮ࢪᅗ Fig. 3 Image of pulse and noise components. (1) ᇶᮏⓗᡭἲ ⤯ᑐ್ ḷၐ㡢ኌࢆ⥺ᙧண ศᯒ㸦LPC㸧ࡍࡿ㸪⥺ᙧண ಀᩘ ṧᕪಙྕࡀᚓࡽࢀࡿ[7]㸬ධຊಙྕࡀࡇࡢࡼ࠺࡞ḷၐ㡢ኌ ࡢሙྜ㸪⥺ᙧண ಀᩘࡣḷモ㒊ศ࣭ኌ㐨ሗ࡛࠶ࡾ㸪ṧᕪ ಙྕࡣ࣓ࣟࢹ㒊ศ࣭ኌᖏሗ࡛࠶ࡿぢ࡞ࡏࡿ㸬๓㡯࡛ グ㏙ࡋࡓࡼ࠺㸪ࣁࢫ࣮࢟࣎ࢫࡣኌᖏࡽ⏕ࡌࡿ㞧㡢ࢆ ྵࡴ㡢ኌ࡛࠶ࡿⅭ㸪ᮏ◊✲࡛ࡣኌᖏሗ࡛࠶ࡿṧᕪಙྕ ฎ⌮࣭ኚࢆຍ࠼ࡓᚋ㸪ඖࡢ㡢ኌࡢ⥺ᙧண ಀᩘྜᡂ ᅗ 4 ࣃࣝࢫᡂศࣀࢬᡂศࡢุᐃࡢ࣓࣮ࢪᅗ ࡍࡿࠕṧᕪኚἲࠖࢆࣁࢫ࣮࢟࣎ࢫࡢᇶᮏⓗᡭἲࡋ Fig. 4 Image of pulse and noise components at the time of [c]࿘Ἴᩘゎᯒࡣࣇ࣮ࣜࢯࣇࢺࠕSound Engineࠖࢆ⏝ࡋࡓ judgment. ⓒ 2013 Information Processing Society of Japan 2 Vol.2013-MUS-99 No.36 2013/5/12 ሗฎ⌮Ꮫ◊✲ሗ࿌ IPSJ SIG Technical Report (3) ṧᕪኚἲ㸰㸦㛫ᘬࡁἲ㸧 ኌᖏሗ࡛࠶ࡿṧᕪಙྕࡣ㸪ࢥࣥࣆ࣮ࣗࢱୖ࠾࠸࡚ ࢧࣥࣉࣜࣥࢢ࿘Ἴᩘ [ d] ࡛⾲⌧ࡉࢀࡓࢹࢪࢱࣝࢹ࣮ࢱ࡛࠶ ⌧≧࡛ࡣಸ㡢㏆ࡢ࿘Ἴᩘᖏᇦࣁࢫ࣮࢟࣎ࢫฎ⌮ࢆ ⾜࠸㸪㡢㧗࿘Ἴᩘᇦࡢ㞧㡢㒊ศࢆඖ㡢ࡢࡲࡲಖᣢࡍࡿ ᪉ἲࡀ᭱㐺ุ᩿ࡋ࡚࠸ࡿ㸬 ࡿ㸬ࡇࡢࢧࣥࣉࣝࡢ୰ࡽࣛࣥࢲ࣒ࢧࣥࣉࣝࢆᢳฟࡋ㸪 ࢹ࣮ࢱ್ࢆ 0 ್⨨ࡁ࠼ࡿ㸦ᅗ 5㸧㸪ࡍ࡞ࢃࡕ㛫ᘬࡁࡍ ࡿࡼࡗ࡚ṧᕪಙྕࡢ࿘ᮇᛶࢆ୍㒊↓ࡃࡋ㸪ఝⓗࣀ ࢬࢆ⏕ࡌࡉࡏࡿᡭἲࡀࡇࡢ㛫ᘬࡁἲ࡛࠶ࡿ㸬㛫ᘬࡁฎ ⌮ࡢࣦ࢙ࣞࣝㄪᩚࡣ㸪యࡢࢹ࣮ࢱᩘࡢఱ㸣ࢆ 0 ್⨨ ࡁ࠼ࡿ㸦㛫ᘬࡁࡍࡿ㸧ࢆᡭືධຊࡼࡾᣦᐃࡍࡿ㸬 ᅗ 6 ศ㞳ࡍࡿ࿘Ἴᩘᖏᇦࢆ♧ࡋࡓᅗ Fig. 6 [ e] Diagram showing a frequency band to separate. ᅗ 7 ᅗ 5 ࢥࣛࢨ࣮ᛂ⏝ἲࡢฎ⌮ࣇ࣮ࣟ Fig. 7 Processing flow of the application equalizer method. 㛫ᘬࡁฎ⌮ࡢ࣓࣮ࢪᅗ Fig. 5 Image of thinning processing. 3.2 ࣁࢫ࣮࢟࣎ࢫࡢ⮬ື⏕ᡂ 3. ṧᕪኚࡢᛂ⏝ἲ 㛫ᘬࡁἲࡢᛂ⏝ࡋ࡚㸪㛫ᘬࡁฎ⌮ࡢࣦ࢙ࣞࣝࢆ⮬ືⓗ ື⏕ᡂἲࠖࢆ⪃ࡋࡓ㸬ࡇࢀࢆḟࡢ㸦1㸧 㸦2㸧 㸦3㸧グ㏙ࡍ 3.1 ࢥࣛࢨ࣮ᢏ⾡ࡢᛂ⏝ ࡿ㸬 Ỵᐃࡍࡿࡼࡾ㸪ࣁࢫ࣮࢟࣎ࢫࢆ⮬ື⏕ᡂࡍࡿࠕ⮬ ࣁࢫ࣮࢟࣎ࢫ࠾࠸࡚᭱ࡶ༴ࡋ࡞ࡅࢀࡤ࡞ࡽ࡞࠸ せ⣲ࡣ㡢㉁ࡢຎ࡛࠶ࡿ㸬ࣁࢫ࣮࢟࣎ࢫࡣඖ᮶㸪㞧㡢ࢆ (1) ྵࡴ㡢ኌ࡛࠶ࡿⅭ㸪ࣁࢫ࣮࢟࣎ࢫ⏕ᡂࢆヨࡳࡿ࠺ࡋ ⮬ື⏕ᡂࡢ㝿ࡢࣦ࢙ࣞࣝỴᐃᣦᶆࡣṧᕪಙྕࡢᖜ ࡚ࡶ㸪వィ࡞㞧㡢ࡢᡂศࡀቑ࠼㸪㡢㉁ࡢຎ⧅ࡀࡗ࡚ࡋ ࡢࣄࢫࢺࢢ࣒ࣛࢆ⏝ࡍࡿ㸬ࡇࢀࡣ୍⯡ࡢேࡢኌẚ㸪 ࡲ࠺㸬ࡋࡓࡀࡗ࡚㸪ࣁࢫ࣮࢟࣎ࢫࡢ㝿ࡣ㡢㉁ಖᣢࡀ ࣁࢫ࣮࢟࣎ࢫࡢṧᕪಙྕࡢᖜࡢศᕸࡀࡁࡃ␗࡞ࡿⅭ ᭱ඃඛ㡯࡞ࡿ㸬 ࡛࠶ࡿ㸬ᮏ◊✲࡛ࡣ᱓⏣ె♸Ặࡢኌࢆ⌮ⓗ࡞ࣁࢫ࣮࢟࣎ 㛫ᘬࡁࣦ࢙ࣞࣝỴᐃᣦᶆ ࡇࡢ㡢㉁ಖᣢࡘ࠸࡚ࡣ㸪ṧᕪኚࢥࣛࢨ࣮ࡢ ࢫࡢࣔࢹࣝࡋ㸪ᖜࡢࣄࢫࢺࢢ࣒ࣛ࠾࠸࡚ࣔࢹࣝ ᢏ⾡ࢆᛂ⏝ࡍࡿࡼࡗ୍࡚ᐃࡢຠᯝࡀ☜ㄆ࡛ࡁࡓ㸬ࡇࢀ ࡢㄗᕪࢆ⟬ฟ㸪ࡇࡢㄗᕪࢆᣦᶆࡋ࡚᭱㐺࡞ฎ⌮ࡢࣦ࢙ࣞ ࡣ୍ᐃࡢ࿘Ἴᩘࡢࡳṧᕪኚࢆ⾜࠸㸪ཎ㡢ࡢୗᆅࢆ୍㒊 ࣝࢆỴᐃࡋ࡚࠸ࡿ㸬ࡲࡓ㸪ኌ࠾࠸࡚ࡢṧᕪಙྕࡢᖜࡢ ṧࡍࡇࡼࡗ࡚㸪㡢㉁ࢆಖᣢࡍࡿ࠸࠺ᡭἲ࡛࠶ࡿ㸦ᅗ ศᕸࡀ 5D\OHLJK ศᕸ>8@ᚑ࠺௬ᐃࡍࡿࡼࡗ࡚㸪ࣃ 6㸧㸬௨ᚋ㸪ࡇࡢᡭἲࢆࠕࢥࣛࢨ࣮ᛂ⏝ἲࠖࡪ㸬 ࣓࣮ࣛࢱ࣮1 ࡘ㸦௨㝆㸪ㄗᕪ Į㸧࡛ㄗᕪࢆ⾲⌧ࡍࡿࡀྍ ࢥࣛࢨ࣮ᛂ⏝ἲ࡛ࡣ㸪ࣁࢫ࣮࢟࣎ࢫ⏝࠸ࡿཎ ⬟࡞ࡗࡓ㸦ᅗ 8㸧㸬ᐇ㝿ࣄࢫࢺࢢ࣒ࣛㄗᕪࢆ⟬ฟࡋࡓ 㡢ࢆ FIR ࣇࣝࢱࡼࡗ࡚タィࡉࢀࡓࣂࣥࢻࣃࢫཬࡧࢫࢺ 㝿ࡢ⤖ᯝࡀᅗ 9 ࡛࠶ࡿ㸬 ࢵࣉࣇࣝࢱࢆ⏝࠸࡚㸪ṧᕪኚࢆ⾜࠺ᖏᇦ㸪ࡑࢀ௨እ ࡢᖏᇦศ㞳ࡍࡿ㸬ศ㞳ࡋࡓ୍᪉ࡢ㡢ኌṧᕪኚ࣭ྜ ᡂࢆ⾜ࡗࡓᚋ㸪ࡇࡢࡘࢆ㊊ࡋྜࢃࡏࡿࡼࡾ㸪୍ᐃ࿘ Ἴᩘࡢࡳṧᕪኚࢆ⾜࠺ࡇࡀྍ⬟࡞ࡿ㸦ᅗ 7㸧㸬ᑦ㸪 [d]ࢧࣥࣉࣜࣥࢢ࿘Ἴᩘࡣ 44100Hz ⓒ 2013 Information Processing Society of Japan [e]࿘Ἴᩘゎᯒࡣࣇ࣮ࣜࢯࣇࢺࠕWave toneࠖࢆ⏝ࡋࡓ 3 Vol.2013-MUS-99 No.36 2013/5/12 ሗฎ⌮Ꮫ◊✲ሗ࿌ IPSJ SIG Technical Report ᅗ 10 ᅗ 8 ṧᕪಙྕࡢࣄࢫࢺࢢ࣒ࣛ Rayleigh ศᕸ 㛫ᘬࡁᗘࢆ 1%ࡎࡘⴠࡋ࡚࠸ࡗࡓ㝿ࡢㄗᕪࡢኚ Fig. 10 Error change when went down by 1% the degree of thinning. (3) ḷၐ༊㛫ࡾฟࡋࡼࡿฎ⌮ไᗘࡢྥୖ Fig. 8 Rayleigh distribution and the histogram of the residual ḷၐࢆ㘓ࡋࡓ㡢ኌࢹ࣮ࢱࡢሙྜ㸪ᜥ⥅ࡂࡢ㝿ࡢᴟࡵ࡚ signals. ᖜࡢᑠࡉ࠸㒊ศࡸ㸪ኌࢆฟࡋ࡚࠸࡞࠸↓㡢ࡢ㒊ศࡀᚲࡎ Ꮡᅾࡍࡿ㸬ࣁࢫ࣮࢟࣎ࢫࢆ⾜࠺㝿㸪ࡇࡢ↓㡢㒊ศࡲ ࡛ฎ⌮ࢆࡍ㸪ணᮇࡏࡠ㞧㡢ࡀຍࡉࢀ࡚ࡋࡲ࠺࡞ 㸦㉥㸧᱓⏣Ặࡢḷၐ㡢ኌ㻌 ࠸ࡗࡓ⢭ᗘຎࡢཎᅉ࡞ࡿ㸬ࡲࡓ㸪➨୍᭱ᑠㄗᕪࡾฟ 㸦㟷㸧ḷၐྜᡂ㡢ኌ㻌 ࡋࡢไᗘࡶⴠࡕ࡚ࡋࡲ࠺㸬ࡑࡇ࡛㸪⮬ື⏕ᡂἲࡢ㝿ࡣ㸪 ᖜࡢᑠࡉ࠸༊㛫ࡸ↓㡢ࡢ༊㛫㸪ḷၐࡢ༊㛫ࢆศ㞳ࡋ㸪 ḷၐࡢ༊㛫ࡢࡳฎ⌮ࢆ⾜࠺ࡼࡗ࡚⢭ᗘࡢྥୖࢆᅗࡿ㸬 ḷၐ༊㛫ࡣ㸪ධຊ㡢ኌࡢᖜࡼࡾ⟬ฟࡍࡿ㸬௵ពࡢ༊ Į 㛫㸦⌧≧࡛ࡣ୍⛊㛫ࢆ 100 ศ㸧ࡢᖜࡢ⤯ᑐ್ࡢᖹᆒࢆ ồࡵ㸪ࡇࡢᖹᆒ್ࡀᑠࡉ࠸㒊ศࢆ↓㡢༊㛫ุ᩿ࡋ㸪ḷၐ ༊㛫ࡢ㛤ጞ࣭⤊Ⅼࡍࡿ㸦ᅗ 11㸧. ࡇ࠺ࡋ࡚⟬ฟࡋࡓྛḷၐ༊㛫ࡈฎ⌮ࢆ⾜࠸㸪᭱⤊ⓗ ᅗ 9 Fig. 9 ṧᕪಙྕࡢࣄࢫࢺࢢ࣒ࣛㄗᕪ⟬ฟ ↓㡢༊㛫ࢆྵࡵ࡚ࡢ༊㛫ࢆ⤖ྜࡍࡿࡼࡾ㸪ฎ⌮ไ Histograms of the residual signal and error calculation. ᗘࡢྥୖࢆᅗࡿ㸬 (2) ➨୍᭱ᑠㄗᕪࡼࡿ᭱㐺ฎ⌮ࣦ࢙ࣞࣝࡢỴᐃ 1࡛グ㏙ࡋࡓᖜࡢࣄࢫࢺࢢ࣒ࣛࡽ⟬ฟࡍࡿ㸪ㄗᕪ Į ࡣ㸪㛫ᘬࡁฎ⌮ࡢࣦ࢙ࣞࣝࢆୖࡆ࡚࠸ࡃ㸪࠶ࡿ୍ᐃࡢᩘ ್ࡲ࡛ࡣୗࡀࡾ㸪ࡑࢀ௨㝆ࡣつ๎࡞ࡩࡿࡲ࠸ࢆࡍࡿ≉ᛶ ࡀ࠶ࡿ㸦ᅗ 10㸧㸬ࡇࡢㄗᕪࡀ᭱ᑠ࡞ࡿึᅇࡢ࣏ࣥࢺࢆ ࠕ➨୍᭱ᑠㄗᕪࠖࡧ㸪ࡇࡢ➨୍᭱ᑠㄗᕪ࡛ฎ⌮ࣦ࢙ࣞ ࣝࢆỴᐃࡍࡿࡼࡾ㸪ᖜࡢࣄࢫࢺࢢ࣒ࣛୖ࡛ࡣࣔࢹࣝ ྠࡌࣁࢫ࣮࢟࣎ࢫࡢ⏕ᡂࡀྍ⬟࡞ࡿ㸬ᑦ㸪➨୍᭱ᑠ ㄗᕪࡢ⟬ฟࡣ㸪᳨ドࡢ⤖ᯝ㸪㛫ᘬࡁࣦ࢙ࣞࣝࢆ 1㸣ࡎࡘୖ ࡆࡿࡓࡧุᐃࢆ⾜࠺᪉ἲࡀ᭱㐺࡛࠶ࡿ㸪ุ᫂ࡋࡓ㸬 ᅗ 11 ḷၐ༊㛫ࡢࡾฟࡋࡢᵝᏊ Fig. 11 Extraction of singing section. 4. ⫈ྲྀᐇ㦂 3 ❶ࡲ࡛㏙ࡓ✀ࡢṧᕪኚἲ㸪ཬࡧᛂ⏝ἲࡼࡗ ࡚ࣁࢫ࣮࢟࣎ࢫࡋࡓ㡢ኌᑐࡍࡿ⫈ྲྀᐇ㦂ࢆ⾜ࡗࡓ㸬 ᐇ㦂ࡢ᪉ἲࡣ㸪ࡘࡢࣁࢫ࣮࢟ྜᡂḷၐ㡢ኌࢆ⪺ࡁẚ ⓒ 2013 Information Processing Society of Japan 4 Vol.2013-MUS-99 No.36 2013/5/12 ሗฎ⌮Ꮫ◊✲ሗ࿌ IPSJ SIG Technical Report 㸪ࡕࡽࡀዲࡳࢆุ᩿ࡉࡏࡿⅬႴዲἲࢆ⏝࠸ࡓ㸬⿕ ᯝ࡛ࡣ㸪ࣀࢬᙉㄪἲࡶከࡃ⚊ࢆ㞟ࡵࡓࡀ㸪ࡇࡢᡭἲࡣࣃ 㦂⪅ࡣ⏨ᛶ㸶ྡ㸪ዪᛶ㸲ྡࡢィ㸯㸰ྡ࡛࠶ࡿ㸬 ࣓࣮ࣛࢱ࣮タᐃࢆᡭືධຊ࡛⾜࠺ᚲせᛶࡀ࠶ࡿⅭ㸪ỗ⏝ᛶ ࡢ㠃ࢆ⪃៖ࡋ࡚ࡶ㸪ࢥࣛࢨ࣮ᛂ⏝ἲ㸩⮬ື⏕ᡂἲࡀ᭱ ⫈ྲྀᐇ㦂⏝ࡋࡓ㡢ኌࡣ ࡶ᭷ຠ࡞ࣁࢫ࣮࢟࣎ࢫࡢᡭẁ࡛࠶ࡿ⪃࠼ࡽࢀࡿ㸬 ࣭ࣀࢬᙉㄪἲ ࣭㛫ᘬࡁἲ ࣭ࢥࣛࢨ࣮ᛂ⏝ἲ 㛫ᘬࡁἲࣀࢬᙉㄪἲࡢẚ㍑ ࣭⮬ື⏕ᡂἲ 12 ࣭ࢥࣛࢨ࣮ᛂ⏝ἲ㸩⮬ື⏕ᡂἲ 10 [ f] ࡢࡘ࡛࠶ࡿ 㸬 ྜィⅬ 8 6 ⫈ྲྀᐇ㦂ࡢ㞟ィ⤖ᯝࡣ௨ୗࡢᅗ 12 ࡢࡼ࠺࡞ࡗࡓ㸬 య㞟ィ⤖ᯝ 4 ྜィⅬ 2 50 0 㛫ᘬࡁἲ 40 ࣀࢬᙉㄪἲ ᅗ 13 㛫ᘬࡁἲࣀࢬᙉㄪἲࡢẚ㍑ 30 Fig. 13 Comparison of noise enhancement and thinning 20 method. 10 ࢥࣛࢨ࣮ᛂ⏝ࡢ᭷↓ࡘ࠸࡚ࡢẚ㍑ 0 12 -10 10 ྜィⅬ 8 6 4 2 ᅗ 12 ⫈ྲྀᐇ㦂ࡢ㞟ィ⤖ᯝ Fig. 12 0 Aggregate results of the listening tests. ↓ࡋ ᭷ࡾ ᅗ 14 ࢥࣛࢨ࣮ᛂ⏝ࡢ᭷↓ࡘ࠸࡚ࡢẚ㍑ ᭱ࡶ⚊ࢆ㞟ࡵࡓࡢࡀ㸪ࢥࣛࢨ࣮ᛂ⏝ἲ⮬ື⏕ᡂἲ Fig. 14 Comparison of the presence or absence of ࢆࡅྜࢃࡏࡓࡶࡢ࡛࠶ࡗࡓ㸬⥆࠸࡚㸪ࣀࢬᙉㄪἲࡶ ࡁ࡞⚊ࢆ㞟ࡵࡓ㸬ࡉࡽྛᡭἲࢆᑐ࡛ẚࡓ⤖ᯝὀ┠ࡋ equalizer application. ࡚࠸ࡃ㸬ᇶᮏⓗ࡞ṧᕪኚἲࡘ࠸࡚ࡣ㸪ࣀࢬᙉㄪἲ ⮬ື⏕ᡂࡢ᭷↓ࡘ࠸࡚ࡢẚ㍑ ᭷⏝ᛶࡀ࠶ࡿ☜ㄆࡉࢀࡓ㸦ᅗ 13㸧㸬ࡇࢀࡣ㸪ࣀࢬᙉㄪ ἲࡀཎ㡢ྵࡲࢀࡿ㡢ኌࢆ⏝ࡋ࡚࠸ࡿࡓࡵ㸪ఝࣀࢬ 10 ࢆⓎ⏕ࡉࡏࡿ㛫ᘬࡁἲࡼࡾࡶ㸪ኚᚋࡢ㡢ኌࡢ⮬↛ᛶࡀಖ 8 ᣢࡉࢀࡓࡓࡵ᥎ ࡉࢀࡿ㸬ࡑࡢࡓࡵ㸪య㞟ィ࡛ࡶࣀ ࢬᙉㄪἲࡣࡁ࡞⚊ࢆ㞟ࡵࡓ㸬ࢥࣛࢨ࣮ᛂ⏝ἲࡢ᭷↓ 6 ࡢ⤖ᯝ㸦ᅗ 14㸧࡛ࡣ㸪᭷ࡾ⚊ࡀ㞟ࡲࡾ㸪ࢥࣛࢨ࣮ᛂ 4 ⏝ἲࡘ࠸࡚᭷⏝ᛶࡀ☜ㄆࡉࢀࡓ㸬⮬ື⏕ᡂἲࡢ᭷↓ࡢ⤖ 2 ᯝ㸦ᅗ 15㸧ࡣ㸪ከᑡ᳨ᐃࢆ⦆ࡵ㸪P0.15 ࡋࡓሙྜ㝈 ྜィⅬ 0 ࡾ㸪᭷⏝ᛶࡀ☜ㄆࡉࢀࡓ㸬⤖ᯝⓗࡘࡢᛂ⏝ἲ㸪ඹ᭷ ↓ࡋ ᭷ࡾ ⏝ᛶࡀ࠶ࡾ㸪ࡲࡓయ㞟ィ࡛㸪ࢥࣛࢨ࣮ᛂ⏝ἲ㸩⮬ື ᅗ 15 ⮬ື⏕ᡂࡢ᭷↓ࡘ࠸࡚ࡢẚ㍑ ⏕ᡂἲࡀ᭱ࡶ⚊ࢆ㞟ࡵࡓࡽ㸪୧⪅ࡢᛂ⏝ἲ㛵ࡋ࡚㸪 㔜ࡡࡅ┦ຠᯝࡶᮇᚅ࡛ࡁࡿุ᫂ࡋࡓ㸬య㞟ィࡢ⤖ Fig. 15 Comparison for the presence of auto-generated. [ f]ඖ㡢࡞ࡿḷၐྜᡂ㡢ኌࡣ㸪ࣇ࣮ࣜࢯࣇࢺࠕ㹓㹒㸿㹓ࠖࢆ⏝ࡋ࡚ไస ⓒ 2013 Information Processing Society of Japan 5 Vol.2013-MUS-99 No.36 2013/5/12 ሗฎ⌮Ꮫ◊✲ሗ࿌ IPSJ SIG Technical Report ࡇࡇ࡛㸪᭱ࡶ⾲ࢆྲྀࡗࡓࢥࣛࢨ࣮ᛂ⏝ἲ㸩⮬ື⏕ᡂἲ ࡢ࿘Ἴᩘゎᯒࡢ⤖ᯝࢆ♧ࡋ࡚࠾ࡃ㸦ᅗ 16㸧㸬⤖ᯝࡽࢃ ㅰ㎡ ◊✲࠶ࡓࡾ㸪༠ຊࢆࡋ࡚࠸ࡓࡔ࠸ࡓ࡚ࡢⓙᵝ ࡿࡼ࠺࿘Ἴᩘᖏᇦࢆᩓᕸࡉࡏࡿᡂຌࡋ࡚࠸ࡿ㸬 ឤㅰࡢពࢆ⾲ࡍࡿ㸬 ඖ㡢 ฎ⌮ᚋࡢ㡢ኌ ᅗ 16 ฎ⌮ᚋࡢ࿘Ἴᩘゎᯒ⤖ᯝ Figure 16 Frequency analysis results after treatment 5. ࠾ࢃࡾ ཧ⪃ᩥ⊩ [1] 㣩ᒇ㸭Ⳮⵦ㸦࠶ࡵࡸ࣭࠶ࡸࡵ㸧: ḷኌྜᡂࢶ࣮ࣝ㹓㹒㸿㹓ࢧ ࣏࣮ࢺ࣮࣌ࢪ (http://utau2008.web.fc2.com/), Kenchan : UTAU ࡢ ࣮࣒࣮࣍࣌ࢪ http://kenchan22.web.fc2.com/homepageright.html [2] ᚨ⏣ᜨ୍, ᾆᆂ୍㑻: ⮬ືᏛ⩦ࡼࡾே㛫ࡢࡼ࠺ḷ࠺㡢 ኌྜᡂࢩࢫࢸ࣒ -Sinsy-, ሗฎ⌮Ꮫ◊✲ሗ࿌, (2012.2.3) [3] 㧗ஂྐ㯢, ⊷⏣ா⏨, ᮧ୍㑻㸸⚟ḟ▮: ᐙᗞ་Ꮫ , ἲ◊, (2010.10.10) [4] ຍ⸨ᆂ㐀, ఀ⸨ᙲ๎: ࢢࣟ࢘ࣝཬࡧࢫࢡ࣮࣒ࣜḷၐࡢྜᡂ ྥࡅࡓ㡢㡪ⓗ≉ᚩࡢศᯒ, ሗฎ⌮Ꮫ◊✲ሗ࿌, (2012), No.14 [5] ຍ⸨ᆂ㐀, ఀ⸨ᙲ๎: 」ᩘḷၐ⪅ࡼࡿࢫࢡ࣮࣒ࣜḷၐ㡢ኌ ࡢ㡢㡪ⓗ≉ᚩࡢศᯒ, ᪥ᮏ㡢㡪Ꮫ⛅Ꮨㅮ₇ㄽᩥ㞟,1-2-9,(2012) [6] ᅵᖌ▱⾜, ᴙᩗⶶ, ∦ᒸⱥᖾ, ⸨ᮌᬸஓ: ṧᕪἼᙧศᯒࡼ ࡿႤኌࡢᐈほⓗホ౯, ⪥㰯⮫ᗋ 86: 7; 1013㹼1018, (1993.3.13) [7] 㟷ᮌ┤ྐ㸸ࢹࢩࢱ࣭ࣝࢧ࢘ࣥࢻฎ⌮ධ㛛, CQ ฟ∧♫ (2006.04) [8] L ࣐ࢮࣝ: ☜⋡࣭⤫ィ࣭ࣛࣥࢲ࣒㐣⛬ (బ⸨ᖹඵヂ), 86㹼90 㡫, ᳃ฟ∧, ᮾி, (1980) ᮏ◊✲࡛ࡣ㸪ḷၐྜᡂ㡢ኌࡢࣁࢫ࣮࢟࣎ࢫࡢ᪉ἲࡘ ࠸࡚ࡘࡢṧᕪኚἲ㸪ཬࡧᛂ⏝ἲࢆᥦࡋ㸪ࡑࢀࡽࡢᡭ ἲ࡛⏕ᡂࡋࡓ㡢ኌᑐࡍࡿ⫈ྲྀᐇ㦂ࢆ⾜ࡗࡓ㸬ὀ┠ࡋࡓ࠸ ࡢࡣ㸪ࡇࡢ⫈ྲྀᐇ㦂࡛⮬ື⏕ᡂἲ᭷⏝ᛶࡀ☜ㄆࡉࢀࡓ ࡛࠶ࡿ㸬ࡇࡢ⮬ື⏕ᡂἲࡣ㸪ᖜࡢࣄࢫࢺࢢ࣒ࣛࡽ⟬ฟ ࡉࢀࡿㄗᕪ Į ࢆඖฎ⌮ࣦ࢙ࣞࣝࢆỴᐃࡋ࡚࠸ࡿ㸬ࡍ࡞ࢃ ࡕ㸪ࡇࡢㄗᕪ Į ࡣࣁࢫ࣮࢟࣎ࢫࡢ㨩ຊࢆᅗࡿ୍ࡘࡢᣦᶆ ࡛࠶ࡿ⪃࠼ࡽࢀ㸪ࡇࢀࡣ┠ᶆࡋ࡚࠸ࡓே㛫ⓗ㨩ຊࢆᅗ ࡿᣦᶆࡶ࡞ࡾᚓࡿྍ⬟ᛶࡀ࠶ࡿ㸬ᚋ㸪ࡉࡽ◊✲ࢆ῝ ࡵ㸪ࡇࡢே㛫ⓗ㨩ຊࢆᅗࡿᣦᶆࡢ☜❧ࢆᅗࡾࡓ࠸㸬 ࡲࡓ㸪ᮏ◊✲࡛ࡣ⏨ᛶࡢྜᡂ㡢ኌࡢࡳࢆᑐ㇟ࣁࢫ࣮࢟ ࣎ࢫࢆ⾜ࡗ࡚ࡁࡓⅭ㸪ዪᛶࡢྜᡂ㡢ኌ㛵ࡋ࡚ࡶࣁࢫ ࣮࢟࣎ࢫࢆྍ⬟ࡍࡿࡢࡀ➨୍ࡢㄢ㢟࡛࠶ࡿ㸬ዪᛶࡢ ྜᡂ㡢ኌࡢࣁࢫ࣮࢟࣎ࢫ㝿ࡋ࡚ࡣ㸪⏨ᛶࡢࡶࡢẚ ࿘Ἴᩘࡀ㧗࠸Ⅼࡸ㸪ࣇ࢛࣐ࣝࣥࢺࡢᙧ≧ࡀ␗࡞ࡿⅬࢆ⪃ ៖ࡋࡓฎ⌮ࡀᚲせ࡛࠶ࡿ᥎ ࡉࢀࡿ㸬 ᭱ᚋ㸪ྜᡂ㡢ኌࡢࣁࢫ࣮࢟࣎ࢫࡢ⏝᪉ἲࡘ࠸ ࡚グ㏙ࡍࡿ㸬ࡇࢀࡣࡶࡕࢁࢇ㸪㡢ᴦไసࡢ㉳⏝ࡀ➨୍ ࡋ࡚ᣲࡆࡽࢀࡿࡀ㸪ࡑࢀ௨እࡶ㸪࠼ࡤ㸪㆙ഛࢩࢫࢸ࣒ ࡢ㆙࿌㡢ኌ࡞㸪┦ᡭᚰ⌮ⓗᙳ㡪ࢆ࠼ࡓ࠸ሙྜࡶ㸪 ࡇࡢࡼ࠺࡞≉Ṧ࡞ྜᡂ㡢ኌࡀຠᯝࢆⓎࡍࡿྍ⬟ᛶࡀ࠶ࡿ㸬 ࡍ࡛㏻ᶵ㛵࡞ࡢࢼ࢘ࣥࢫ࡛㉳⏝ࡉࢀ࡚࠸ࡿࡼ࠺㸪 ᚋ㸪♫ࡢᵝࠎ࡞ሙ㠃࡛ࡢྜᡂ㡢ኌࡢ⏝ࡀቑຍࡍࡿ ࡀணࡉࢀࡿ㸬ࡑࡢࡓࡵ㸪ྜᡂ㡢ኌே㛫ⓗ㨩ຊࡸ≉Ṧ࡞ ຠᯝࢆຍࡍࡿᢏ⾡ࡢᚲせᛶࡣᚲࡎ㸪㧗ࡲࡗ࡚࠸ࡃ࡛࠶ࢁ ࠺㸬ࡑ࠺࡞ࡗࡓ㸪ᮏ◊✲ࡢᡂᯝࡀ㈉⊩࡛ࡁࡿࡼ࠺㸪ࡇ ࢀࡽࡶ◊✲ࢆ⥆ࡅ㸪ࡉࡽ࡞ࡿⓎぢ㸪㐍ᒎ⮳ࢀࢀࡤⰋ࠸ ⪃࠼࡚࠸ࡿ㸬 ⓒ 2013 Information Processing Society of Japan 6