FPGA Implementation of Dynamic Quantile Tracking based Noise Estimation for Speech Enhancement
Bittu Kumar, Ashwini Kumar Varma · Journal of Engineering Science and Technology Review · 2023
In this paper, the architecture of Dynamic Quantile Tracking (DQT) for estimating the noise signal has been proposed for Spartan-6 DSP Field Programmable Gate Array (FPGA) device.It is simulated on MATLAB/Simulink environment using Xilinx System Generator (XSG) blocks for speech enhancement application and implemented on Spartan-6 SC6SLX45 FPGA board in real-time mode.The hardware resources for the proposed architecture were obtained in terms of Flip Flops, Logic Slices, LUTs, Bonded IOBs, DSP48A1s and RAMB18E1s.The performance of the proposed architecture has been evaluated and analyzed with the variation of different speech-to-noise ratio levels and varying types of noises.Also, their results are compared with other methods like cascaded median and modified cascaded median methods.All results are evaluated and measured in teams of PESQ score with variations of SNR level and various noises and also presented in waveforms and spectrogram.In terms of PESQ score, it is observed that the enhanced speech for the proposed model is a good quality speech signal compared to other method's enhanced speech and noisy speech signals.In terms of hardware, it is observed that based on the utilization of resources, the proposed model acquires only 30% of the resources of the Spartan-6 SC6SLX45 FPGA board during real-time processing.