Title:
音声認識性能の予測システム、学習モデルの構築方法、及び、音声認識性能の予測方法
Document Type and Number:
Japanese Patent JP7426686
Kind Code:
B2
Abstract:
To provide a prediction system for voice recognition performance that precisely predicts voice recognition performance while saving the trouble to measure an impulse response in a user environment and the costs for the measurement.SOLUTION: A prediction system 100 for voice recognition performance comprises a learning model 11 that has been machine-learnt to output a prediction value PV of voice recognition performance in a space in which a reverberation voice is obtained when a feature quantity FV as a value based upon the reverberation voice W is input.SELECTED DRAWING: Figure 1
Inventors:
Takahiro Fukumori
Takanobu Nishiura
Takanobu Nishiura
Application Number:
JP2019114876A
Publication Date:
February 02, 2024
Filing Date:
June 20, 2019
Export Citation:
Assignee:
School corporation Ritsumeikan
International Classes:
G10L15/01
Other References:
郭 他,"ノンリファレンス特徴量を用いた自然発話音声認識の性能推定の検討",日本音響学会2016年秋季研究発表会講演論文集CD-ROM,2016年08月31日,pp.121-124
福森隆寛 他,"PESQと室内音響指標を用いた雑音・残響指標NRSR-PAに基づく雑音・残響下音声認識性能の予測",電子情報通信学会論文誌D,2015年03月01日,Vol.J98-D, No.3,pp.343-352
福森隆寛 他,"PESQと室内音響指標を用いた雑音・残響指標NRSR-PAに基づく雑音・残響下音声認識性能の予測",電子情報通信学会論文誌D,2015年03月01日,Vol.J98-D, No.3,pp.343-352
Attorney, Agent or Firm:
Hiroshi Sakamoto