基于GMM统计参数和SVM的说话人辨认研究

Text-Independent Speaker Identification Based on GMM Statistical Parameters and SVM

支持向量机(SVM)是以统计学习理论为基础,解决模式识别问题的有力工具,但是它训练算法复杂,难以处理大量样本,限制了其在说话人识别方面的使用。针对这个问题,提出了一种基于GMM(高斯混合模型)统计参数和SVM的说话人辨认系统,以GMM模型的统计参数来训练SVM说话人辨认模型,有效解决了大样本数据下SVM模型的训练问题。实验表明,该方法有良好的效果,并且与倒谱加权方法结合后,可以增强系统的健壮性,进一步提高系统的识别率。

Support vector machine （SVM） based on statistical learning theory is a powerful tool for pattern recognition problems. However, since the training algorithm is too complex to accept large quantitise of trainging data, the use for speaker recognition is limited. A system based on Gaussian mixture model（GMM） statistical parameters and SVM for text-independent speaker identification is proposed to aim at this problem. The SVM speaker model is trained from the parameters of GMM, which is an efficient method on large-scale training data. The results of experiments showed that the system has good performance. Especially when combined with cepstral coefficient weighting method, the system＇s robustness and performance are further improved.