The acoustic modeling is established on the Duration Distribution Based Markov Model (DDBHMM).But the recognition rate slowed down obviously for the vocabulary's doubly incensement

in view of this problem

we used multithreading in the longest time-consume compute part to optimize the recognition.Moreover

the vertex examination is under consideration

then the record module

feature-extraction module

wave-show module and output-show module were designed

finally

we integrated all the modules with interface

and developed an independent Uyghur continuous speech real-time recognition system based on DDBHMM

related tests have been done on the system.