The SOPC Implementation Of Blind Source Separation Of Convolution Speeches

In the real environment, speech signal is often interferenced by other signals or noises. Therefore, speech enhancement becomes an important part of speech signal processing. Blind source separation can recover the source signals in the case that the source signals and the parameters of transmission channel are unknown. In order to implement speech separation hardware, this paper design convolution algorithm for blind separation of speech with the system on a programmable chip (SOPC) technologySince the frequency domain convolution algorithm has better performance than the time domain one in speech separation. The thesis uses frequency domain blind speech separation algorithm to perform speech separation. A semiblind algorithm based on negative entropy maximization (CMN) separation and the energy alignment is specifically used for separation. The SOPC design is mainly based on Embedded Development Kit provided by Xilinx. At the same time, the PowerPC processor which is embedded in the FPGA is used as the core processor. The hardware platform is built by using free IPs existed a or provided by the third party vendors, including DDR memory, GPIO control, RS232 serial, AC₉7 audio controller and so on. In the part of the software design, the C language is used as the programming language to achieve the separation of speeches. The results compared with Matlab show that the mixed speeches can be separated.