Speaker identification using Neural Networks on an FPGA

In this work we present the implementation of a speech recognition system. This implementation was made using a neural network as recognizer element and Linear Prediction Coding (LPC) as vector input for the net. This neural network of four layers was implemented on a board Nexys 2 of Digilent which has a Field Programmable Gate Array (FPGA) of XILINX Spartan-3E. The system was trained and tested with 10 users. Each user recorded 6 samples of their voice with different intonation, leading to 60 samples, where 50% of them were used for training and the rest for testing. With this system a recognition rate of 98% was accomplished.