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

国際会議 Interspeech 2017 報告:音声明瞭性

国際会議 Interspeech 2017 報告:音声明瞭性

第120回音声言語情報処理研究会・第188回音楽情報科学研究会 共催研究会で使用したスライド(修正版)

9f4daf3767cef6224907d172ac505867?s=128

Katsuhiko Yamamoto

February 21, 2018
Tweet

Transcript

  1. Ի੠໌ྎੑ ࢁຊ ࠀ඙ ࿨Վࢁେֶ γεςϜ޻ֶݚڀՊ ത࢜ޙظ՝ఔ ೔ຊֶज़ৼڵձ ಛผݚڀһʢDC2ʣ Mail :

    yamamoto.katsuhiko@g.wakayama-u.jp Web : https://sites.google.com/site/kyama0321/ ୈ120ճԻ੠ݴޠ৘ใॲཧݚڀձ ୈ188ճԻָ৘ใՊֶݚڀձ ڞ࠵ݚڀձ ࠃࡍձٞʢInterspeech 2017ʣใࠂ
  2. Ի੠໌ྎੑʹؔ͢Δൃද 1. F. B. Gelderblom, T. V. Tronstad, and E.

    M. Viggen, “Subjective intelligibility of deep neural network-based speech enhancement,” in Proc. Interspeech 2017, pp. 1968– 1972, 2017. 2. A. H. Andersen, J. M. de Haan, Z. H. Tan, and J. Jensen, “On the use of band importance weighting in the short-time objective intelligibility measure,” in Proc. Interspeech 2017, pp. 2963–2967, 2017. 3. C. Spille and B. T. Meyer, “Listening in the dips: Comparing relevant features for speech recognition in humans and machines,” in Proc. Interspeech 2017, pp. 2968–2972, 2017. 21 Feb. 18 Ի੠໌ྎੑʢ࿨Վࢁେ ࢁຊʣ 2
  3. Ի੠໌ྎੑʹؔ͢Δࢦඪ Ի੠ྃղ౓ n Ի੠ʢ୯ޠ΍จষʣ͕ͲΕ͚ͩ໌ྎʹฉ͖औΕΔ͔ n େ͖̎͘छྨͷධՁํ๏ Ø ओ؍ධՁɿௌऔ࣮ݧͷਖ਼౴཰ Ø ٬؍ධՁɿԻ੠ྃղ౓༧ଌ๏ʹΑͬͯಘΒΕΔ਺஋

    21 Feb. 18 Ի੠໌ྎੑʢ࿨Վࢁେ ࢁຊʣ 3 STOI (Taal et al., 2011) n ؆୯ͳ৴߸ॲཧͰɼओ؍ධՁ݁Ռͱͷ૬͕ؔߴ͍ n Ի੠ڧௐ෼໺Ͱͷ٬؍ධՁख๏ͱͯ͠ਓؾ C. H. Taal, R. C. Hendriks, R. Heusdens, and J. Jensen, “An algorithm for intelligibility prediction of time- frequency weighted noisy speech,” IEEE Trans. Audio. Speech. Lang. Processing, Vol. 19, No. 7, pp. 2125–2136, 2011.
  4. Gelderblom et al. (2017) ֓ཁ n DNNϕʔεͷԻ੠ڧௐॲཧख๏ Ø Xu et

    al. (2015) ͷ֦ுϞσϧ Ø STOIͷ਺஋͕࠷େʹͳΔΑ͏ʹ࠷దԽ 21 Feb. 18 Ի੠໌ྎੑʢ࿨Վࢁେ ࢁຊʣ 4 n ݁Ռ Ø STOIɿڧௐԻ੠ͷ໌ྎੑ͸ڧௐલΑΓվળ Ø ௌऔ࣮ݧɿશͯͷඃݧऀʹ͓͍ͯɼڧௐԻ੠ͷԻ੠ྃղ౓ ͕௿ԼʢڧௐલΑΓྼԽʣ Ø STOIͷ༧ଌ݁Ռ͸ௌऔ࣮ݧͷ݁Ռͱໃ६ n Ի੠໌ྎੑͷධՁ Ø ٬؍ධՁɿSTOI Ø ओ؍ධՁɿௌऔ࣮ݧ
  5. Andersen et al. (2017) ֓ཁ n STOIͷԻ੠ྃղ౓ͷ༧ଌਫ਼౓ͷվળ Ø όϯυॏΈ෇͚ؔ਺ʢband importance

    functions; BIFsʣ Ø ෳ਺ͷσʔλϕʔεͷओ؍ධՁ݁ՌʹରԠՄೳʢ≈globalʣ ͳBIFΛ࣮ݧ͔Βܾఆ͢Δɽ n ݁Ռ Ø શͯͷ৚݅Ͱྑ͍༧ଌਫ਼౓Λࣔ͢BIF͸ݟ͔ͭΒͣ Ø ൺֱతޡࠩͷখ͍͞BIFͷ݁Ռ͸ɼΦϦδφϧͷSTOIͰ࢖༻ ͞Ε͍ͯΔBIFʢ܎਺͕શͯۉҰʣͷ݁Ռʹ͍ۙ Ø BIFΛ༻͍ͨSTOIͷ༧ଌਫ਼౓ͷվળ͸೉͍͠ 21 Feb. 18 Ի੠໌ྎੑʢ࿨Վࢁେ ࢁຊʣ 5
  6. Spille & Meyer (2017) n DNNϕʔεͷԻ੠ೝࣝثΛԠ༻ͨ͠Ի੠ྃղ౓༧ଌ๏ Ø ೖྗɿର਺ϝϧεϖΫτϧʢ11ϑϨʔϜ෼ʣ Ø ग़ྗɿ̏ͭ૊ԻૉʢτϥΠϑΥϯʣ

    ֓ཁ n ֶशͨ͠৚݅ͷԻ੠ྃղ౓͸ߴਫ਼౓Ͱ༧ଌՄೳ Ø ఆৗࡶԻʢSSNʣ Ø ඇఆৗࡶԻʢSAM-SSNʣ n ਂ૚ֶशͷख๏ʢLRPʣΛ༻͍ͯ໌ྎੑΛ෼ੳՄೳ Ø τϥΠϑΥϯͷࣝผʹରͯ͠ͷೖྗͷߩݙ౓Λදࣔ Ø ໌ྎੑͷ൑அ͕୯७ͳᮢ஋൑ఆͰແ͍͜ͱΛࣔࠦ 21 Feb. 18 Ի੠໌ྎੑʢ࿨Վࢁେ ࢁຊʣ 6