论文部分内容阅读
针对现有关键词识别系统采用很难用硬件电路准确描述的连续隐马尔可夫模型CHMM作为识别模型,提出用离散隐马尔可夫模型DHMM作为系统的识别模型,研究了适用于硬件实现的状态机端点检测算法,并通过引入VQ矢量量化模块来保证离散关键词识别系统的识别率和识别速度;根据关键词训练模型,分析所采集语音信息中是否存在指定的关键词并进行准确识别。实验结果表明,该算法在便于硬件实现的基础上,具有良好的识别率和实时性,为关键词识别系统的FPGA硬件电路实现研究提供了参考。
Aiming at the existing key word recognition system using the continuous hidden Markov model CHMM, which is difficult to describe accurately by the hardware circuit, as a recognition model, a discrete hidden Markov model DHMM is proposed as a recognition model of the system, and the state suitable for hardware implementation Machine end-point detection algorithm is introduced, and the VQ vector quantization module is introduced to ensure the recognition rate and recognition speed of the discrete keyword recognition system. According to the keyword training model, the key word is analyzed to determine whether there is a specific keyword in the collected speech information and to identify accurately. The experimental results show that the algorithm has good recognition rate and real-time on the basis of easy hardware implementation, and provides a reference for the realization of FPGA hardware circuit of keyword recognition system.