音声スペクトルの時間変化強調による明瞭性改善手法の提案  [in Japanese] Improvement Method of Speech Intelligibility by Enhancement of Spectral Variation  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

It is hard to listen to an announcement in a noisy environments such as in a train while running. In this paper, to improve speech intelligibility in a noisy environment, we focused on the time variation of speech spectral representing spectral flux from the analysis result which clear speech has larger spectral flux than normal speech. Therefore, we propose speech enhancement method of spectral variation to improve the sound quality evaluation value of STOI. In the evaluation results of STOI score, enhanced speech by proposed method improved more than original speech in all the SNR conditions (SNR = -5, -10, -15 dB). Also, the significant difference in the dangerous rate of 5% was found in SNR = -10, -15 dB.

It is hard to listen to an announcement in a noisy environments such as in a train while running. In this paper, to improve speech intelligibility in a noisy environment, we focused on the time variation of speech spectral representing spectral flux from the analysis result which clear speech has larger spectral flux than normal speech. Therefore, we propose speech enhancement method of spectral variation to improve the sound quality evaluation value of STOI. In the evaluation results of STOI score, enhanced speech by proposed method improved more than original speech in all the SNR conditions (SNR = -5, -10, -15 dB). Also, the significant difference in the dangerous rate of 5% was found in SNR = -10, -15 dB.

Journal

  • The Japanese Journal of the Institute of Industrial Applications Engineers

    The Japanese Journal of the Institute of Industrial Applications Engineers 6(1), 51-54, 2018

    The Institute of Industrial Applications Engineers

Codes

  • NII Article ID (NAID)
    130006507532
  • Text Lang
    JPN
  • ISSN
    2189-373X
  • NDL Article ID
    029502703
  • NDL Call No.
    Z74-J567
  • Data Source
    NDL  J-STAGE 
Page Top