Design of a speaker recognition system

Abstract

In recent years, various cryptosystems have appeared from the laboratory to the market. Human nature as a unique characteristic for different people has been utilized in these various cryptosystems to do the security work. For example, fingerprint unlocking has been commonly used in cell phones and controlling access to the door. Iris recognition is also a new and developing technology used in an upscale environment. Another technology called speaker recognition uses the speech of different people to achieve identification and verification in the security system. In this report, a speaker recognition system is built to recognize which person is speaking in a record of speech and whether this speaker is in the database of authority or not. There are two main parts of this system: The first part is the speech feature extraction system. The technology called Mel-frequency cepstrum coefficients (MFCC) is used to extract the unique features in the speech for different speakers. The features represent for different people will be stored in the database. The second one is speech feature matching system, using vector quantization (VQ) to compare the features of the testing speaker with the all the speaker features in the database and determine whether this speaker is in the database or not. After these two parts, a complete speaker recognition system will be constructed and can be used to achieve identification and verification in a security system.