Speech recognition and hardware accelerator design based on hummingbird core

Abstract

In recent years, with the rapid development of artificial intelligence, speech recognition has irreplaceable advantages in human-computer interaction. In human-computer interaction, we hope that speech recognition has the characteristics of fast recognition speed, high recognition accuracy, and strong flexibility. This paper develops a hardware accelerator based on a programmable gate array, in which the nucleic acid detection systems utilize voice to complete machine control. We use a traditional BP neural network to train the parameters of the speech recognition model, and it is successfully deployed on FPGA to realize speech recognition. The acceleration effect is significantly improved, realizing that the cycles used before acceleration are almost four times those used after acceleration.