摘要
针对实时性问题提出了一种以FPGA为硬件平台的说话人识别系统解决方案。该方案以MFCC为语音特征,采用了基于矢量量化的说话人识别算法。系统主要包括语音信号采集、端点检测、特征提取和识别判断4个部分。经测试证明,该系统完成了设计所需的基本功能。在实验室条件下,当系统时钟为50 MHz时,完成一次4码的识别耗时15.932 ms,对12码的识别率为93.3%。
For real-time problem, this paper presents a speaker's voice recognition system solution that makes the FPGA as the hardware platform. The system consists of four parts: signal acquisition, endpoint detection, feature extraction and identification. The experiment results show that the time-consuming is 15.932 ms on the 4 codebooks and 50 MHz-clock system, the identifica- tion rate is 93.3% on the 12 codebooks system. This kind of design improves the system's recognition speed, which is an effective program to solve the real-time problem.
出处
《电子技术应用》
北大核心
2012年第11期16-18,21,共4页
Application of Electronic Technique