Research Of Sound Localization And Speech Enhancement And Implementation Based On SOPC

Sound localization and speech enhancement system are widely applied in fieldsof telecommunication, military security, artificial intelligence, industrial manufacture,etc. Embedded system based upon microphone array is moving towards largecalculation, high integration, high resolution and low complexity. Being of thistendency, SOPC System for sound localization and speech enhancement has beendesigned and achieved based on SOPC architecture.Sound localization algorithm will be applied in this research. This paperanalyzes phase weighting method (PHAT)’s considerable ameliorations upongeneralized cross-correlation (GCC)’s properties. It validates steered response power-phase weighting (SRP-PHAT) method’s excellent time delay estimation performance.SRP-PHAT method distinguishes itself in highlighted peak value, high-resolutionsound localization and outstanding performance in the case of multiple sound sources.CFRC algorithm is applied in this thesis to reduce sound localization’s largecalculation quantity.Speech enhancement algorithm will also be discussed in this thesis. In terms ofbalance performance and complexity, adaptive beamforming is employed asmicrophone array speech enhancement algorithm. Analysis upon the performance ofadaptive beamforming’s general model-generalized sidelobe canceller (GSC), showsthat GSC excels in coherent noise elimination, but fails in noncoherent noiseelimination. Through adding frequency domain wiener filtering to GSC’s nonadaptivebranch, it performs well in inhibition of both the coherent and the noncoherent noise.To improve adaptive algorithm’s convergence speed, both NLMS a daptive algorithmand subband filter bank are adopted. Simulation proves that the subband GSCexceeds3dB SNR compared with full band GSC. To verify this algorithm’sengineering performance, simulations have been taken on floating point numbers,fixed-point numbers and compromise in the process of project implementation. Itturns out that the algorithm has practical applications.The entire systematic algorithm is designed and implemented within XilinxXCV5LX110T FPGA, including data cache module, endpoint detection module,subband GSC module, Codec controller, and embedded CPU. It is embedded CPU’sinternal softwares that implements system initialization, human-computer interaction, and sound localization algorithm. Subband GSC module includes the time dela ycompensation, subband filter bank, wiener filtering, NLMS filtering, etc. And eachmodule is designed, programmed, optimized, multi-level simulated, and board-leveldebugged.The hardware platform is designed. Data acquisition of high precision, multi-channel and synchronization is achieved to meet the requirements of array signalcollection. The system is built upon Xilinx V5-LXT FPGA ML505evaluationplatform with LCD adopted as positioning display module and audio codec AD1981as result output device. Referring to3D acoustic detection system, an array with fivemics is adopted in this paper.In the laboratory environment, the system is tested and proved to be effective.