论文部分内容阅读
基于互信息理论的语音识别方法不仅考虑了语音信号的时变分布特征，并且考虑了语音信号的统计分布特征，能有效地提高同类模式的凝聚度，减少非同类模式间的耦合性，在语音识别实验和实际应用中反映出良好的识别精度和很高的运行效率，与其它方法相比更适合嵌入式系统的语音识别应用。本文提出了一种互信息估计的非线性搜索算法，这一算法能够有效地处理语音信号时变分布特征的非线性波动，进一步提高语音模式互信息匹配的精度。
The speech recognition method based on mutual information theory not only considers the time-varying distribution of speech signals, but also takes into account the statistical distribution of speech signals, which can effectively increase the cohesion of similar modes and reduce the coupling between different modes, Recognition experiments and practical applications reflect a good recognition accuracy and high operating efficiency, compared with other methods is more suitable for embedded speech recognition applications. This paper presents a nonlinear search algorithm for mutual information estimation, which can effectively deal with the nonlinear fluctuations of the time-varying distribution of speech signals and further improve the accuracy of mutual matching of speech patterns.