Research Of Speech Recognition Used In The Appliance Control

Speech recognition is a hot topic in the field of speech signal processing. It is a technology that makes computers to understand people's voice commands and carry out simple actions. This human-computer communication method is a nature way for users. Last two to three decades, speech recognition have a wide range of applications in many fields like, industrial, military, transportation, medical industry, especially in the computer, information processing, communications and electronic systems, automatic control. Today, speech recognition products take more and more important place in the human-computer interaction applications.Recently, speech recognition technology applied in embedded system is mainly simple voice command recognition. It makes the work operated by hand can be easily accomplished with voice. It can be used in many home appliances, toys, smart phones and other portable mobile devices.Compared to PC, the embedded speech recognition system whose computing speed and storage capacity is limited, has its own special features. The embedded speech recognition system has many advantages, such as small size, high reliability, low power consumption, less investment, low price, portability is good, can support mobile operations. These advantages of embedded speech recognition system make it be widely used in many areas.Speech recognition can be divided into Speaker-Independent recognition and Speaker-Dependent recognition. For our system, the users are all the members of the family. Therefore, this system should be Speaker-Independent. According to embedded system's computational speed, and storage space, it is necessary to appropriately improve the sophisticated speech recognition for efficiency and accuracy.The work included in this thesis can be divided into the following 4 Parts:Chapter 1 introduces the basic conceptions and principles involving in the speech re cognition system, which includes the pretreatment of speech signal, the technology of the characteristic parameter extracting, template matching and technology of training of model。Chapter 2 implements the endpoint detection algorithm, jointly with the sub-word processing. Three characteristic extractions and two pattern recognition warping methods are studied. These algorithms are implemented in MATLAB.Chapter 3 implements our speech recognition system on a small vocabulary using ZLG EASYARM2200. All the algorithms which are used in the software are implemented in C program language. And design circuits for voice capture, voltage bias adjust and other related circuit.Chapter 4 shows experiments with different values of the key parameter and the recognition rate. Then this thesis discusses the influence of key parameter on the recognition rate.