The Design And Implementation Of The System Under The Colored Background Noise For Speech Enhancement

With the rapid development of communication technology, speech enhancement is becoming one of the most important technical means in the communications field, which is playing an increasingly important role in speech recognition, noise reduction and speech coding, etc, and becomes the hot topic in the voice processing field for nearly 30 years. It targets on improving the language signal to noise ratio and voice quality, and then improving the comfort and intelligibility of the speech signal. It has a very important practical significance.Digital voice signal model is the basis of speech signal processing. The accuracy of the model will directly affect the subsequent processing of the speech signal. This thesis has built an all-pole model, which is simple in structure and easy to implement. Speech enhancement algorithm focuses on the parameter estimation of speech signal and noise made model, model parameters is the basis of the Kalman-filter algorithm. Noise parameter estimation is mainly detecting the effectiveness of the voice through the VAD algorithm, and directly estimating the silence segments of the signal through LPC autocorrelation method. Then some common parameter estimation algorithms are used at the same time, such as maximum likelihood function method, Burg algorithm, etc. However, at low SNR, the error will be increasing when these methods are used, and the algorithm will become unstable. Therefore, on this basis, this thesis proposes a parameter estimated algorithm which is based on high-order accumulation. Pure voice signal and noise signal with high order accumulation is equal according to the characteristics of high order accumulation. Based on this fact, we can put through the AR model through the high order accumulation, and then gets the MYW equation. The main methods for solving the MYW equations are auxiliary variable method, the LMS algorithm. Through the improved LMS algorithm, this thesis presents the conjugate gradient algorithm. The experiments prove that the effect is similar with these algorithms, but the operation can simplify. Finally, speech enhancement is in the way of Kalman-filter. Through the simulation results analysis, the conjugate gradient algorithm which combined with the high-order accumulation can achieve a good speech enhancements effect. It improves the scope enhancement algorithms while simplifies the operations.Finally, in order to verify the feasibility and real-time of the algorithm, this thesis designs the hardware platform. The processor is the chip of TMS320VC5402, audio signal acquisition chip is TLV320AIC23 B and controller is STC89LE58 RD +. This platform not only costs savings, but also improves the efficiency.