FPGA Implementation of Radial Basis Function and Recurrent Neural Network for Speech Recognition

Abstract

Neural networks can catch needlessly deterministic nonlinear and for machine learning non-parametric models it has merits such as fast, highly accurate computation when compared to other probable computing. In this paper, feasibility of Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) and Radial Basis Function Neural Network (RBFNN) for speech recognition is studied. The stochastic incline descent (SGD) method is applied to update the parameters of RBFNN and the temporal classification method is used for training the LSTM-RNN, whereas FSM is used to reluctant the hardware resource usage. The automatic proposal mechanization is done by Simulink and Xilinx is used to verify the Verilog code and it is implemented in FPGA-SPARTAN-6..