CAJ | 학술논문

本文介绍了连续语音识别中噪音鲁棒性方法的现状，分析传统动态范围调整方法在连续语音识别中导致的特征曲线中的峰值不匹配问题，提出了新的特征曲线调整算法，并为算法的参数设定增加了限制条件。本文详细地论述了算法的步骤以及算法中参数的确定。通过理论分析和实验验证，证明了提出的算法在连续语音特征调整中具有很好的性能，提高了识别精度。
본문개소료련속어음식별중조음로봉성방법적현상，분석전통동태범위조정방법재련속어음식별중도치적특정곡선중적봉치불필배문제，제출료신적특정곡선조정산법，병위산법적삼수설정증가료한제조건。본문상세지논술료산법적보취이급산법중삼수적학정。통과이론분석화실험험증，증명료제출적산법재련속어음특정조정중구유흔호적성능，제고료식별정도。
The noise robust methods in continuous speech recognition is introduced, the peaks mismatching problem which exist in continuous speech feature adjustment by using conventional dynamic range adjustment is analyzed. The new feature adjustment algorithm is proposed and some restricted conditions to get parameters are added. The step and the parameter setting of the algorithm paper are discussed in details. By theoretical analysis and experimental verifica-tion, the performance of the algorithm in continuous speech feature adjustment is proved. The recognition accuracy is improved.