Robust Computer Voice Recognition Using Improved MFCC Algorithm

This paper presents an improved MFCC technique for voice recognition and analysis by the computer. The computer records the voice pattern of the speaker during training phase. The parameters for the voice are extracted, graphed and inserted into a database. During the testing phase, a person speaks into the microphone, and the computer analyzes it. Two pattern matching algorithms are used in the recognition mode. Using the conventional MFCC algorithm, the analyzed data is slow (0.12 ms) and inaccurate (66% accuracy). This paper explores the possibility of a new MFCC algorithm that is capable of over 80% accuracy with less than 0.1ms CPU time taken for processing. This algorithm could be used in security devices that use voice recognition technology for identification.