Abstract: Response rate of recognition is a key factor for a speech recognition system as well as recognition correct rate and noise robust property. This paper improves the conventional DTW algorithm’s structure, and fastens the running speed in real-time speaker identification. The advantage of the new algorithm is more obvious with the increment of speeches. Experiments show that the method can increase the operation speed by 1.5 times.

Key words: speaker identification, Mel-frequency Cepstral Coefficient(MFCC), Dynamic Time Warping(DTW)

摘要： 识别正确率和抗噪性能是语音识别的研究重点，而识别响应速度也是决定系统实用化的关键。文章改进了传统的动态时间弯折算法结构，将其应用于实时说话人辨识系统中，极大地提高了系统运行速度，随着待识别语音数目的增多，该算法优势更加明显。实验表明，在不影响系统识别率的情况下，该方法使系统的运行速度平均提高了1.5倍。

关键词: 说话人辨识, 美尔倒谱系数, 动态时间弯折

CLC Number: 

• TP301.6