A study of duration in continuous speech recognition based on DDBHMM

DDBHMM solved the defects of traditional HMM.Based on DDBHMM,the problem of how to effectivelyutilize the duration information is studied in detail.The approach on estimating the duration distribution is introducedfirstly?then the data file is classified according to the speak rate.The recognition experiment shows that,the durationinformation behaves best on the data of low speak rate,behaves normal on the data of medium speak rate and has littleeffect on the data of fast speak rate.Therefore,the most importance of duration is that by it the more accurate statesegmentation point could be obtained and then the recognition rate can be improved.At the same time,the robustnessof the system to speaking rate is improved with the employment of the duration information.Furthermore,the methodof classified duration and normalized duration is also put forward and studied in detail,it shows that both of the twomethod can improve the effect.In order to study the dependency between the duration,the method of using the Bigramof the duration is proposed and analyzed.At last,the approach of post processing duration is studied,it shows thatonly based on DDBHMM,and utilizing the duration information synchronously in the recognition process,then theperformance can be improved greatly.