The Design And Realization Of Wireless Acoustic System Faced To Smart-House

Convenient man-machine interaction makes speech controller a hot point.The mainstream products have powerful feature yet they need networking support or are expensive. I try to design a cheap and excellent one without relying on internet which aims to specific speaker, isolated words recognition system for small amount of vocabularies.The system filters out the noise generated by recording device on the basis of its frequency spectrum and finds the starting point and terminal point according to energy. The system extracts MFCC parameter of speech and compresses it to fixed size matrix using VQ algorithm. Through fixed point realization of process which consists of a mass of similar calculations can reduce count time greatly and losses little precision. The system uses look-up table method to take the place of time consuming trigonometry operations and manages speech feature libraries by FAT32 file system.The test result is that average accuracy of 40 words selected randomly in 200 test comes up to 100% which achieves the target request. The times needed in recognising speech of 2 seconds are about 2.18 seconds which are less than requested 3 seconds. The transmitter is about ten meters far from the receiver in indoor, the test result is that 250 codes are totally received accurately.