基于语音识别的说话人身份辨识系统

Speaker's Identity Recognizing System Based on Speech Recognition

提出一种在小字库孤立语音条件下，集成语音识别与说话人的识别技术，并进行说话人身份代码（密码）识别、认证．利用语音信号的短时分析技术进行孤立词的单元分割，采用临界带特征矢量作为语音信号特征，分析了经典语音识别算法——动态时间规整算法，提出了对语音模板各帧加权的改进方法．为提高识别响应速度，研究了多门限多轮次的判决方法，在增加多套模板、提高识别率的情况下，降低了系统的响应时间．

The paper presents a way for recognizing the speaker's identity or security number string by using speech and speaker recognition technology. The transient analysis technique of speech signal is applied to separate the isolated words, and the critical bands vector is used to describe the features of speech signal. This paper also analyzes a classic speech recognition algorithm——dynamic time warping algorithm and makes some improvement on the weighted vioce frame templates. In order to improve the recognition rate and shorten the system response time, a multi threshold and multi turn decision method is provided.