论文研究了汉语小词汇表语音识别算法的基本原理,提出了具有鲁棒性的两级端点检测语音识别技术,在语音信号采集时,根据过零率、短时能量对数据进行提取并压缩,采用了多模板匹配算法识别。硬件采用51内核单片机,用较少的存储空间和计算空间实现语音数据处理,不需要额外的器件。实验用20个字的汉语小词汇量系统进行了测试,识别成功率大于90%,显示该算法比通常采用的算法性能更好。 The thesis studies the basic principle of Chinese small vocabulary speech recognition algorithm and proposes a robust two-stage endpoint detection speech recognition technology. At the time of speech signal acquisition, the data is extracted and compressed according to zero-crossing rate and short-time energy, Using a multi-template matching algorithm to identify. Hardware uses 51-core microcontroller, with less storage space and computing space for voice data processing, no additional devices. The experiment uses a 20-word Chinese small vocabulary system to test. The recognition success rate is more than 90%, which shows that the algorithm performs better than the commonly used algorithm.