Standalone ASR systems are gradually getting mature,and are used in many fields,in our lives.But ASR alone could only help us in a shadow level.Having made machine capable of transcribing,human naturally hope that it should be able to understand with or even communicate with human.Spoken dialog system is exactly a mixture of speech related techniques that takes heavy responsibility of enabling human-machine communication.A modern dialog system is compose of several modules.Any module's failure could highly possibly to cause the entire dialog to fail.This subject takes several aspects to make spoken dialog system more stable.To deal with the inevitability error of ASR,directed speech error correction method is proposed in this thesis.To rapidly expand domains that spoken dialog system supports,massive data from original domain should be used to help expanding,domain customization should be supported,spoken language understanding that supports domain transfer and customization is studied in this thesis.Some slot values are really essential to spoken dialog system but they are usually too long to be recognized correctly in one try,slot value correction method in dialog is designed in this thesis.These aspects improve spoken dialog system's acceptability on different circumstances or different problems. |