基于HMM和聚类的英语语音识别混合算法

Hybrid English speech recognition algorithm#$NBSbased on HMM and cluster

摘要:

对于具有大量特征数据和复杂发音变化的英语语音,与单词相比,在隐马尔可夫模型(HMM)中存在更多问题,例如维特比算法的复杂度计算和高斯混合模型中的概率分布问题。为了实现基于HMM和聚类的独立于说话人的英语语音识别系统,提出了用于降低语音特征参数维数的分段均值算法、聚类交叉分组算法和HMM分组算法的组合形式。实验结果表明,与单个HMM模型相比,该算法不仅提高了英语语音的识别率近3%,而且提高系统的识别速度20.1%。

Abstract:

For English sounds with a large number of characteristic data and complex pronunciation variations, there are more problems in hidden markov models (HMM) than words, such as complexity calculation of wittby algorithm and probability distribution in gaussian mixture model.In order to realize the speech-independent English speech recognition system based on HMM and clustering, a combination of segmented mean algorithm, clustering crossover grouping algorithm and HMM grouping algorithm is proposed to reduce the dimension of speech feature parameters.Experimental results show that compared with the single HMM model, the algorithm not only improves the recognition rate of English speech 3%, but also improves the recognition speed of the system 20.1%.