CAJ | 학술논문

　　针对语音识别中固定帧率特征提取方法没有充分考虑语音频谱变化特性、噪声鲁棒性差的问题，提出了一种基于自适应加权和的变帧率方法用于特征提取，并在固定音频检索系统中进行实验，在信噪比为20 dB的情况下，与固定帧率的特征提取方法相比，系统检出率提高了近4%。实验表明，该方法在降低噪声影响，提高固定音频检索性能方面是有效的。
　　침대어음식별중고정정솔특정제취방법몰유충분고필어음빈보변화특성、조성로봉성차적문제，제출료일충기우자괄응가권화적변정솔방법용우특정제취，병재고정음빈검색계통중진행실험，재신조비위20 dB적정황하，여고정정솔적특정제취방법상비，계통검출솔제고료근4%。실험표명，해방법재강저조성영향，제고고정음빈검색성능방면시유효적。
Given the problems of fixed frame rate feature extraction method which did not fully consider the changes in charac-teristics of speech spectrum, and had a poor noise robustness, this paper proposes a variable frame rate method based on adap-tive weighted-sum for speech feature extraction, and experiments are conducted with a specific audio retrieval system. Com-pared to the fixed frame rate feature extraction method, under the signal noise ratio of 20 dB, the system detection rate increases nearly 4%. Experimental results show that this method is effective to reduce noise, and to improve the performance of fixed au-dio retrieval.