【摘要】 介绍了用离散隐马尔可夫模型(DHMM)构造孤立词语音识别系统的过程,重点针对软件实现中的问题重新推导了Baum Welch算法的重估公式,引入一种抗噪性能很好的特征参数:过零率与峰值幅度特征,将该特征与DHMM结合用于孤立词识别系统。结果表明,此系统训练时收敛很快并且识别效果好。更多还原

【Abstract】 This paper described the whole process of building a speech recognition system of isolated words using discrete HMM(Hidden Markov Model) and derived the reestimate formula of Baum-Welch algorithm again with respect to the problems in the process of software implement. This paper also introduced a kind of feature parameter which has excellent anti-noise capability: ZCPA feature. Combining this kind of feature with DHMM can form a recognition system of isolated words. The experiment showed that it can converge rapidly in system training process and the recognition results are satisfactory.更多还原