管制指令特征参数提取研究

在管制指令语音识别系统的研究中,语音的特征参数提取方法是影响系统识别率的关键因素之一。在Matlab平台上基于MATLAB GUI技术,设计完成HMM语音训练识别模型的用户操作界面,实现特征参数提取方法的选择、语音信号的训练、识别操作及结果输出和波形显示。通过实例对比分析,特定人和非特定人情况下,LPCC和MFCC两种特征参数提取方法在训练、识别时间和识别率上的差异。结果表明,特定人时LPCC参数的识别优势明显;非特定人时,MFCC参数的识别效果和效率更好。

Research on the Control Instruction Feature Parameter Extraction

In the control instruction speech recognition system, the method of feature parameter extraction of speech was one of the key factors of the system. On the matlab platform based on matlab GUI technology, the user interface of the HMM speech training and recognition model was designed and it could be used to choose the method of feature parameter extraction,the training and recognition operation of speech signal, and to output the results and display the waveform. In the situation of specific person and non-person, comparative analysis the differences of the training, recognition time and recognition rate of between two methods of feature parameter extraction of LPCC and MFCC by examples. The results show that LPCC parameters has more advantages than MPCC parameters while the speaker is a specific person ; MPCC parameters has better recognition effect and efficiency while speakers are non-persons.