Speech recognition method base on memristor

Abstract

The invention discloses a speech recognition method base on a memristor; the method comprises the following steps: selecting a memristor device and building a memristor nerve network unit; building ato-be-trained memristor speech recognition network; inputting speech training signals to the to-be-trained memristor speech recognition network, and returning a recognition result to the FPGA; using the FPGA to calculate deviations between the recognition result and the speed training signals, and training the to-be-trained memristor speech recognition network, thus obtaining the needed memristorspeech recognition network; inputting a to-be-recognized speed signal so as to obtain a speech signal content text. The advantages are that the method can reduce precision and preparation difficulty and time influences on hardware, and the hardware can realize a higher level multilayer convolution nerve network; the method can effectively combine the speech recognition with the memristor, and usesa batched connection mode to reduce the hardware crosstalk interferences, thus greatly improving the speech recognition accuracy.

Description

technical field [0001] The invention relates to the technical field of speech recognition based on neural network, in particular, a speech recognition method based on memristor. Background technique [0002] Memristor, the full name of memory resistance (Memristor), is a circuit device that represents the relationship between magnetic flux and charge. Memristor has the dimension of resistance, but unlike resistance, the resistance value of memristor is determined by the charge flowing through it. Therefore, by measuring the resistance value of the memristor, the amount of charge flowing through it can be known, thereby having the effect of memorizing charge. The emergence of nanometer memristive devices is expected to realize non-volatile random access memory. Moreover, the integration, power consumption, and read / write speed of memristor-based random access memory are superior to those of traditional random access memory. Additionally, memristors are the best way to impl...

AI Technical Summary

Problems solved by technology

[0004] Although scholars have implemented the application of neural networks on memristor arrays, the neural networks implemented are relatively basic, and there is still a distance between the methods and practical applications.

At the same time, the problem of severe noise caused by hardware crosstalk has not been solved well. There is still a blank area in the connection between software and hardware. There is little involvement in the application of speech recognition in memristor.

Images