An attempt to reproduce WaveNet-based text-to-speech synthesis

MACHINE LEARNING Meetup KANSAI #2 on June 15. https://mlm-kansai.connpass.com/event/88458/

音声サンプル:
- https://r9y9.github.io/wavenet_vocoder/
- https://r9y9.github.io/blog/2018/05/20/tacotron2/

Transcript

1. 8BWF/FUח״׷窟鎘涸갈㡮さ䧭ח䮋䨌׃׋鑧 2018/06/15 Ryuichi Yamamoto @ LINE Corp. MACHINE LEARNING Meetup KANSAI 1
2. 5FYUUPTQFFDI
   TZOUIFTJT 2 5FYU
3. 荈䊹稱➜ with my own text-to-speech (TTS) )J
   NZ
   OBNF
   JT
   3ZVJDIJ
   :BNBNPUP
   *N
   B
   TPGUXBSF
   FOHJOFFS
   XPSLJOH
   BU
   -*/&
   DPSQPSBUJPO
   ,ZPUP 3
4. WaveNet: 3遤תה׭ • (PPHMF
   %FFQ.JOE
   ח״׏ג涪僇 • 傀㶷ך405"׾㣐ֹֻ♳㔐׷荈搫䚍ך넝ְㅷ颵 • 堣唒㷕统ك٦أך窟鎘涸갈㡮さ䧭ٌرٕ 4 https://deepmind.com/blog/wavenet-generative-model-raw-audio/
5. WaveNet: 岚䕎ٖكٕך荈䊹㔐䌓㘗欰䧭ٌرٕ 5
6. ֶ⠗ִ׃׋ְֿה • 갈㡮さ䧭UFYUUPTQFFDIכ(16אד׮㨣׭׵׸׷ • 8BWF/FUח״׷넝ㅷ颵ז갈㡮さ䧭׮갹䓸׸ל⦐➂ד׮〳腉 6
7. ➭䩛岀הך嫰鯰 (1/2) 7 Text: Scientists at the CERN laboratory say they have discovered a new particle Deep Voice3 [Ping ‘18]. (trained on LJSpeech, w/o WN) Tacotron 2 [Shen; ‘18]. (trained on LJSpeech , w/ WN) Tacotron 2 [Shen; ’18]. (trained on proprietary corpus , w/ WN)
8. ➭䩛岀הך嫰鯰 (2/2) 8 Text: Generative adversarial network or variational auto-encoder. Deep Voice3 [Ping ‘18]. (trained on LJSpeech, w/o WN) Tacotron 2 [Shen; ‘18]. (trained on LJSpeech , w/ WN) Tacotron 2 [Shen; ’18]. (trained on proprietary corpus , w/ WN)
9. Ꟛ涪ך酅⩎ 9
10. GPUך侧ָ駈׶זְ㉏겗 10 https://arxiv.org/abs/1712.05884
11. GPUך侧ָ駈׶זְ㉏겗ח㼎ׅ׷鍑瘶 11 • 孡さ – 〴 (59
    5Jד갹䓸׶ת׃׋ – ָծ湫䠬דְֽ׷ה䙼׏׋ךדծװ׶ֹת׃׋
12. 㷕统ח֮ת׶ח儗꟦ַַָ׷㉏겗 • ⡦傈ַַ׏׋ך٥٥٥ – رٌ갈㡮ٖكٕך넝ㅷ颵ח麦ׅ׷חכ鹈꟦ֻ׵ְכ䗳銲L – 傈ד׮ TPGUNBY׾⢪ִל
    ׉׸ז׶ךㅷ颵חכז׷ • 鎘皾׃גְ׷꟦זח׃גְ׋ך٥٥٥ –
    ˘Ī˘ ˘ 12
13. 䱿锷ח֮ת׶ח儗꟦ַַָ׷㉏겗 • 猱ך갈㡮ךさ䧭ח ⴓ PO
    (PPHMF
    $PMBCPSBUPSZ –
    ˘Ī˘ ˘
    13
14. ֶ⠗ִ׃׋ַ׏׋ֿה • 갈㡮さ䧭כ(16
    אַ׵㨣׭׵׸תׅ 8BWF/FUכ㼰׃鳞ְדָׅ
    • 8BWF/FU
    דכꬊ䌢ח넝ㅷ颵ז갈㡮さ䧭ָ〳腉 – 剑鵚ך،فٗ٦ثכ سً؎ٝ濼陎׉׿זחְ׵זְךדծٖحخزٓ؎ 14
15. 剑䖓ח 15 Thank you for coming to the machine learning meetup Kansai on June 15!
16. References • <%FFQ.JOE
    ˎ>
    8BWF/FU
    "
    (FOFSBUJWF
    .PEFM
    GPS
    3BX
    "VEJP IUUQTEFFQNJOEDPNCMPHXBWFOFUHFOFSBUJWF NPEFMSBXBVEJP • <WBO
    EFO
    0PSE
    ˏD>
    "BSPO
    WBO
    EFO
    0PSE
    4BOEFS
    %JFMFNBO
    )FJHB ;FO
    FU
    BM
    8BWF/FU
    "
    (FOFSBUJWF
    .PEFM
    GPS
    3BX
    "VEJP
    BS9JW
    4FQ
     • <1JOH
    ˎ>
    8FJ
    1JOH
    ,BJOBO 1FOH
    "OESFX
    (JCJBOTLZ
    FU
    BM
    ˑ%FFQ
    7PJDF
    
    4DBMJOH
    5FYUUP4QFFDI
    XJUI
    $POWPMVUJPOBM
    4FRVFODF
    -FBSOJOH˒
    1SPD
    PG
    *$-3
     • <4IFO
    ˏ>
    +POBUIBO
    4IFO
    3VPNJOH 1BOH
    3PO
    +
    8FJTT
    FU
    BM
    /BUVSBM
    554
    4ZOUIFTJT
    CZ
    $POEJUJPOJOH
    8BWF/FU
    PO
    .FM
    4QFDUSPHSBN
    1SFEJDUJPOT
    
    1SPD
    PG
    *$"441
     16