砂時計型ニューラルネットワークの多段化によるLSPパラメータ圧縮特性の改善  [in Japanese] Improvement of Compression Characteristic of LSP Parameters by Cascading Sandglass Type Neural Network  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

我々は,多段接続5 層砂時計型ニューラルネットワーク(CSNN(NL5))を用いた日本語5母音のLSPパラメータの情報圧縮と特徴抽出を行う手法を提案した.男性話者1名による日本語5母音の発話資料を用いてCSNN(NL5) の有効性を実証した.これにより,1) CSNN(NL5) により2次に圧縮されたパラメータの分布は,第1,第2フォルマントの分布と類似した分布を示すこと,2) CSNN(NL5) は,圧縮したLSPパラメータを音声合成に使用できる精度で復元できることを明らかにした.We proposed a new scheme that derives the characteristics of Japanese five vowels out of LSP parameters by compressing information in terms of cascaded five-layer-sandglass-type neural network (CSNN(NL5)). We have verified the ability of CSNN(NL5) by using five vowels pronounced by a male speaker . The followings were clarified, 1) the distribution of LSP parameters compressed by CSNN(NL5) is similar to the distribution of F1-F2 formants, 2) CSNN(NL5) can reproduce the LSP parameters from the compressed parameters usable for speech synthesis.

We proposed a new scheme that derives the characteristics of Japanese five vowels out of LSP parameters by compressing information in terms of cascaded five-layer-sandglass-type neural network (CSNN (NL5)). We have verified the ability of CSNN (NL5) by using five vowels pronounced by a male speaker. The followings were clarified, 1) the distribution of LSP parameters compressed by CSNN (NL5) is similar to the distribution of F_1-F_2 formants, 2) CSNN (NL5) can reproduce the LSP parameters from the compressed parameters usable for speech synthesis.

Journal

  • IPSJ journal

    IPSJ journal 46(3), 845-848, 2005-03-15

    Information Processing Society of Japan (IPSJ)

References:  4

Codes

  • NII Article ID (NAID)
    110002768586
  • NII NACSIS-CAT ID (NCID)
    AN00116647
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • ISSN
    1882-7764
  • NDL Article ID
    7280182
  • NDL Source Classification
    ZM13(科学技術--科学技術一般--データ処理・計算機)
  • NDL Call No.
    Z14-741
  • Data Source
    CJP  NDL  NII-ELS  IPSJ 
Page Top