A Keyword Recognition System Based on Hybrid Compression Convolutional Neural Network

Abstract

The invention provides a keyword recognition system based on a hybrid compressed convolutional neural network, which belongs to the technical field of calculation, calculation and counting. The system includes an analog-to-digital conversion module, a feature extraction module, and a hybrid compression convolutional neural network module, which respectively perform high-order residual quantization on the input value of the convolutional neural network, and perform fine-grained progressive quantization and activation value on the weight value of the neural network. The low-bit quantization is performed on the convolution kernel, and then the convolution kernel is pruned based on the joint evaluation strategy of the front and rear stages, so as to reduce the amount of parameters of the network and the size of the model. The parameter size of the neural network is reduced by quantization, and the calculation amount of the network is reduced by pruning, so as to achieve the purpose of optimizing the network.

Description

technical field [0001] The invention discloses a keyword recognition system based on a hybrid compressed convolutional neural network, relates to the compression optimization of the convolutional neural network, and belongs to the technical field of calculation, calculation and counting. Background technique [0002] Since the performance of the deep neural network far exceeds the traditional machine learning algorithm, deep learning technology is constantly advancing in many fields, and has broad application prospects in the fields of wearable devices, robots, and smart homes. It has achieved much better performance than in the past. Among them, the convolutional neural network is a very important implementation method, but the scale of the neural network is huge, and there are more parameters and calculations than traditional methods. A large-scale deep neural network requires a large amount of storage resources to support its calculations. [0003] The large storage requ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to address the shortcomings of the above-mentioned background technology, and propose a keyword recognition system based on a hybrid compressed convolutional neural network to meet the needs of neural networks deployed on mobile terminals and portable devices, and to achieve low power consumption, high Accurately complete the keyword recognition task, and solve the technical problem that the existing voice keyword recognition system is difficult to implement in mobile terminals and portable devices

Images