Based On The Improved Mfcc Parameters Of The Application Of Speech Recognition System

With the rapid development of modern computer technology, the human-computer interact get more and more frequently, we need to complete many function that impossible for human features with the aid of machines. How to make the machines execute human command faster and better has gradually become a popular research content. The widely used speech recognition technology can make machines understand human voice, which saving time and efficiency greatly. This technology not only from the lab to large industrial facilities, but also into the human’s daily life, with a wide variety of embedded products.The paper describes the development of speech recognition, introduces the theory and technology. We focused analysis several classic algorithms such as the endpoint detection, the parameters and template matching, then select the appropriate algorithm to build a speech recognition system according to the feature of system. We improve the defect that the traditional MFCC parameters are variable by the pitch of speakers, through smoothing the signal spectrum envelop first, then calculation the MFCC parameters. The SMFCC parameters improve the stability and reduce the impact of pitch. We also simplify the calculation of DTW, improving the response time of system.Considering the limitation of the voice signal transmission distance, we establish a reliable network connection through the TCP transport protocol, to transmit the identified voice command to the remote machines to execute. This method can improve the performance of the speech signal, and increase the market value of speech recognition technology.After building the corresponding hardware and software environment, we write and debug the program to test in the Tiny6410development board. We choose20voices of a male and a female as the template separately, and ask5males and5females to test the performance of speech recognition and transmission, then record the recongnition rate and response time. The test shows that the recognition rate and the response time of the system are effective by the improved characteristic parameters and the DTW algorithm.The system still exist defects to be refined and improved, which need the further research to improve.