内窥镜自动定位语音识别系统.docxVIP

ComputerEngineeringandApplications计算机工程与应用2014，50（8）207

内窥镜自动定位语音识别系统

马宁，陈晓冬，李亚楠，尹青云，汪毅，郁道银

MANing,CHENXiaodong,LIYanan,YINQingyun,WANGYi,YUDaoyin

天津大学精密仪器与光电子工程学院，天津300072

CollegeofPrecisionInstrumentsOpto-electronicsEngineering,TianjinUniversity,Tianjin300072,China

MANing,CHENXiaodong,LIYanan,etal.Speechrecognitionforendoscopicautomaticpositioningsystem.Com-puterEngineeringandApplications,2014,50（8）：207-210.

Abstract：Anovelsystemforminimallyinvasivesurgeryispresentedinthispaper.ThesystemutilizesanEndoscopicAutomaticPositioner（EAP）controlledbyspeechrecognitionenginetoimplementtheclampinganddynamicalpositioningofthelaparoscope.ThemotioninstructionsoftheEAParetransformedfromvoicecommandsofspecificdoctorrecog-nizedbyspeakerdependentspeechrecognitionalgorithmnamedDynamicTimeWarping（DTW）.TheDTWrecognizesparticularcommandsandrejectsirrelevantitemsbyenhancingtheperformanceofthereferencetemplate.AnARM-coreembeddedplatformisdesignedtoruntheDTWonWindowsCEoperatingsystem.Andonthatbasis,theperformanceofDTWisdemonstratedby1250groupsofexperimentsfrom10individualspeakers.Comparedwiththetraditionalalgo-rithm,theenhancedalgorithmcanimprovetherecognitionrateby3.16%andshortenthetimeofcalculationby51%.Theresultsdemonstratetheavailabilityoftheenhancedalgorithmanditsabilitytosatisfytherealtimerequirementinembed-dedsystem.

Keywords：endoscopic;dynamictimewarping;referencetemplate;speakerdependent;embeddedsystem

摘要：提出一种基于特定人的内窥镜自动定位语音识别系统，通过识别特定医生的语音控制口令实现内窥镜的定位，为手持内窥镜操作提供更加智能化的解决方案。在识别算法上提出了参考模板归一化平均的动态时间规划（NormalizedAverage-DynamicTimeWarping，NA-DTW）算法，可获得更高的识别率，系统以片上WindowsCE操作系统和ARM作为系统的软硬件平台。实验通过对10个不同测试人的共1250组测试数据进行识别检测，NA-DTW算法与传统DTW算法相比，识别率从96.6%提高到99.76%，运算时间从469ms缩短到241ms。验证了NA-DTW算法可以完成基于特定人、孤立词的语音识别功能，并满足嵌入式系统中