A Short Speech Speaker Recognition Methods And Applications

After decades of development, audio processing technology has made a lot of breakthrough, many audio technologies like speech recognition technology and speech synthesis technology has reached the level of large-scale applications. As a natural way of interaction, the popularity of audio technology will facilitate each person’s daily life.Speaker recognition is an important branch of speech processing. In the middle of1990s, especially after the application of Gaussian mixture model in the speech field, the speaker recognition technology achieved great development and improvement. Under the condition of adequate input speaker voice and in a quiet ideal environment, the speaker recognition can achieve a high recognition rate, even can beyond the level of ordinary human identification, and it can basically meet the requirements of the practical applications. However, in realistic scenarios, due to the particularity and complexity of the open environment,there are many confounding factors in the system, like the noise, channel, the training speech length and the test speech length. The performance of the system will always degrade significantly and can’t obtain the result of recognition in the laboratory environment, which directly affect the large-scale popularization and application of speaker recognition technology.The short-term test speech is a factor that affects the performance of speaker recognition. Due to the complexity and limitation of realistic application scene or application requirements, the original speech signal will contain too few of the characteristics of the speaker’s personality, which lead to the dramatic degradation of the recognition performance. Against the influence of short-term utterance, this paper proposes a speaker recognition algorithm based on the selection use of common features. This method makes feature selection in speaker training and test process, reducing the features those may lead to wrong decisions, decreasing the risk and improve the recognition rate.This paper firstly works on a complete speaker recognition system based on Gaussian mixture model. Then it goes into the short-term speech problem and some of the short-term utterance speaker recognition algorithms, and points out the existing problems. By the analysis of the shortcomings of those algorithms, the author were inspired to think out a simpler and more effective short-term speaker recognition algorithm, which improve the current algorithm. In this paper, the new identification algorithm is applied to a toy robot system, and the test results show that the performance has been improved significantly. The new algorithm can effectively identify the short-term speech of0.25seconds than the existing traditional algorithms, making the toy robot more smart and entertaining.