摘要
通过对硬件平台特点的研究,采用了多帧同步搜索算法、马氏距离并行内积化和并行指令集的技术对语音识别和说话人识别进行了优化,识别速度提高了3倍。在P4 1.4GHz的机器上,可以同时完成120路的说话人识别任务。
A novel set efficient algorithms is introduced. With the study of hardware platform of personal computer, cache optimization, changing Mahalanobis distance to parallel inner product and utilizing the single instruction multiple data stream instruction set are presented. With these methods, the speed of recognition of both speech recognition and speaker identification increase about 3 times.
出处
《计算机工程》
CAS
CSCD
北大核心
2004年第15期1-3,共3页
Computer Engineering
基金
国家自然科学基金资助项目(60272016)
关键词
语音识别
说话人识别
快速算法
并行内积化
Speech recognition
Speaker identification
Efficient algorithm
Parallel inner product
