一种语音识别的可定制云计算方法  

A Customizable Cloud Computing Method of Speech Recognition

作　　者：张巍[1] 贾玉辉[1] 张志楠[1] 

机构地区：[1]中国海洋大学信息科学与工程学院,山东青岛266100

出　　处：《中国海洋大学学报（自然科学版）》2014年第1期112-117,共6页Periodical of Ocean University of China

基　　金："语音库自动构建方法和语音极小化标注"研究项目(ZR2012FM016)资助

摘　　要：以云的方式提供语音识别服务是1种新的趋势。当前市场上的语音云都是面向通用领域的,针对这一现状,提出了1种面向特定领域,可定制的,可满足用户个性化需求的云计算方案。首先介绍了语音识别的MapReduce过程,接着描述了基于Sector/Sphere的语音识别系统组成及语音识别的Sphere处理流程,并给出了语音识别系统的云计算架构,然后探讨了语音识别的可定制实现,最后进行实验评估,证明云计算架构的语音识别系统与传统C/S架构的语音识别系统相比更具优势,且具有一定的可扩展性。Speech-recognition services in the cloud computing＇s way is a new trend. Voice clouds compu- ting on the current market are used in general field. Under this situation,we put forward a cloud compu- ting solution which was used in specific field. This solution was customizable, so it could meet the needs of individual users. At first, we introduced the MapReduce process of speech recognition. Then, described the composition of the speech recognition system which based on Sector/Sphere and processing flow of speech recognition, and gave the cloud computing Architecture of the speech recognition system. We also explored the customizability method of speech recognition. At last, through a test evaluation, we proved that compared with traditional c/s architecture speech recognition system,the cloud computing architecture speech system had more advantages, and it had a certain degree of scalability.

关 键 词：云计算 语音识别 SECTOR SPHERE MAPREDUCE 可定制 

分 类 号：TN912.34[电子电信—通信与信息系统]