System Design And Implementation Of Speech Enhancement Based On DSP

The research of speech enhancement techniques plays a crucial role to the quality enhancement of speech communication. Speech enhancement technique is to separate useful speech signals from noise with some signal processing techniques so as to increase the signal-noise ratio, to enhance the intelligibility of noise-corrupted speech signals and to decrease the auditory tiredness of humans ears.Broadband Gauss white noise and single-frequency colored noise are two kinds of very common noises in short-wave serial wireless communication. The purpose of this dissertation is to propose a solution to effectively eliminate them based on theoretical research and simulation result analysis.There are two logical parts of this dissertation.The first part is algorithm research and simulation. First, it studies two broadband speech enhancement techniques such as MMSE, algorithm based on perceptual auditory masking effects and then proposes subband filtering algorithm based on critical band. The focus is algorithm research and simulation result analysis. Besides, MMSE is chosen to be finally implemented on hardware platform according to the needs of fixed-point programming and real-time implementation. Secondly, it studies the characteristics and speech enhancement principle of single-frequency colored noise, does the simulation and further proposes an improved solution.The second part concentrates on the construction of DSP communication platform and algorithm migration, including the communication mechanism between these two chips, system DSP/BIOS configuration and C programming and optimization. In this project, TI TMS320VC5509 and C8051 F410 MCU are used to build the hardware platform. Up to now, both the broadband and the single-frequency speech enhancement algorithms have run successfully at real time on the hardware platform.