Font Size: a A A

The Research Of Speech Identification Based On Wireless Communication And Its Software Development

Posted on:2006-02-01Degree:MasterType:Thesis
Country:ChinaCandidate:L J LiuFull Text:PDF
GTID:2168360155964676Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Speech recognition is one of the important contents of the speech signal processing, and how to increase the preciseness in speech recognition is the mainly intention of the speech recognition research. In the present paper, we firstly focus on improving the preciseness and robust of the speech recognition through precisely representing the speech signal by suitable speech features. On the other hand, speaker recognition is a special application of the speech recognition and, has a widely foreground. In this thesis, we also discuss the algorithms of the speaker recognition. What's more, with the increasingly development of the wireless communication, the speech recognition based on wireless communication is becoming more importance. Owing to the multi-path fading of the channel, there exists the new requirement for speech recognition when the speech signals go through the wireless channel. So, here we try to deal with how to use computers to simulate the environment of transferring the speech signal in real communication channel, with which one can find the suitable methods of increasing the preciseness in speech recognition based on wireless communication. Our major contributions are as follows: (1) We analyze the fading principle of wireless signal communication, and present a multi-path fading model of the signal in the process of the transmission. In other words, if the delay expanding of the channel is longer than the sign period of transmitted signal, the signal transmitting will lead to frequency selective fading and ISI (Inter-symbol interference). Because of the stationary stochastic process of wireless channel, we choose a Gauss process to simulate the multi-path fading of channel. Indeed, the simulation provides a real environment of speech signal communication, and can help us to find a suitable method in speech recognition based on wireless communication. (2) We discuss some major algorithms of speech recognition, and provide a new speech recognition method to increase the preciseness rate, i.e., we firstly extract the features of the signal by MVDR, and regard the features as the new observation, and then transform the features into the new features by ICA Algorithm. Finally, based on the corresponding features, ICA is used for recognizing the speech. The experiments demonstrate that the present algorithm dealing with the noise speech recognition based on MVDR spectrum estimation and ICA algorithm is valid. (3) Based on the limitation of the speaker recognition algorithms, in this thesis, we indicate that the effect of speaker recognition through support vector machines algorithm is better than that through Gaussian mixture model (GMM), but the efficiency of the SVM algorithm is worse than the GMM algorithm. So, we present a novel way, namely Optimal Support Vector Machines (OSVM), which is used for reducing the number of support vector machines so that it can improve the efficiency of speech recognition.
Keywords/Search Tags:Speech recognition, Support vector machine, Wireless communication, Minimum variance distortionless response, Independent component analysis
PDF Full Text Request
Related items