音声・音響信号を対象としたブラインド音源分離  [in Japanese] Blind Source Separation for Speech and Acoustic Signals  [in Japanese]

Search this Article

Author(s)

    • 猿渡 洋 SARUWATARI Hiroshi
    • 奈良先端科学技術大学院大学 情報科学研究科 Graduate School of Information Science, Nara Institute of Science and Technology

Abstract

本稿では,音声・音響信号を対象としたブラインド音源分離(BSS)技術について概説する.ブラインド音源分離技術は,その信号処理形態の面より周波数領域独立成分分析(FDICA)と時間領域独立成分分析(TDICA)とに大別される.まず初めに,FDICA及びTDICAに基づく従来のBSSについて簡単な解説を行い,その問題点を述べる.特に,FDICAにおける帯域分割数と狭帯域信号間の独立性に関するトレードオフ問題について詳細な説明を行い,音声分離実験の結果を紹介する.次に,近年我々が提案している多段ICA法を紹介し,それらが残響環境における信号分離性能の面にて優れていることを示す.

This paper reviews several techniques of blind source separation (BSS) used for speech and acoustic signals. These techniques can be classified into two main groups in terms of the type of processing, i.e., frequency-domain independent component analysis (FDICA) and time-domain independent component analysis (TDICA). First, conventional BSS methods based on FDICA and TDIGA are briefly reviewed and their drawbacks are discussed. In particular, the trade-off problem between the number of subbands and independence among narrow-band signals in FDICA is explained in detail with the extensive results of speech-separation experiments. Next, we introduce a multistage ICA method proposed in our recent works by emphasizing the effectiveness regarding the signal-separation performances under a reverberant condition.

Journal

  • Technical report of IEICE. DSP

    Technical report of IEICE. DSP 101(667), 59-66, 2002-02-25

    The Institute of Electronics, Information and Communication Engineers

References:  22

Cited by:  8

Codes

  • NII Article ID (NAID)
    110003280612
  • NII NACSIS-CAT ID (NCID)
    AN10060786
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • ISSN
    09135685
  • NDL Article ID
    6143482
  • NDL Source Classification
    ZN33(科学技術--電気工学・電気機械工業--電子工学・電気通信)
  • NDL Call No.
    Z16-940
  • Data Source
    CJP  CJPref  NDL  NII-ELS 
Page Top