Speech Recognition under Noisy Environments using Spectral Subtraction with Time Dimensional Smoothing

2 Citations 5 References

Bibliographic Information

Other Title
  • スペクトルサブトラクションと時間方向スムージングを用いた雑音環境下音声認識

Search this article

Description

One of the problems on the application of speech recognition system in the real world is degradation of the performance by surrounding noises. We propose to use the spectral subtraction (SS) with time dimensional smoothing. In this method, which called SS-SMT in this paper, we can take the difference of phases between speech and noise into account. Experimental results show the improvement of 27% recognition rate from the method without SS, 13% from the original SS. When using SS-SMT in training acoustic models in noisy environments, the variance of noises is taken into account. This method achieves the further improvement of 11% recognition rate from SS-SMT with original acoustic models.

Journal

  • IEICE technical report. Speech

    IEICE technical report. Speech 98 (106), 31-37, 1998-06-12

    The Institute of Electronics, Information and Communication Engineers

Citations (2)*help

See more

References(5)*help

See more

Keywords

Details 詳細情報について

  • CRID
    1573668927130521344
  • NII Article ID
    110003296943
  • NII Book ID
    AN10013221
  • Text Lang
    ja
  • Data Source
    • CiNii Articles

Report a problem

Back to top