基于小波神经网络的嵌入式语音识别系统设计  

Design of an Embedded Speech Recognition System Based on Wavelet Neural Network

作　　者：陈龙飞[1] CHEN Longfei(Shanghai Normal University,Shanghai 201418,China)

机构地区：[1]上海师范大学,上海201418

出　　处：《电声技术》2024年第7期25-27,共3页Audio Engineering

摘　　要：针对嵌入式语音识别中的资源受限问题,提出一种基于小波神经网络的轻量化识别方案。该方案利用小波变换提取语音信号的时频域特征,并结合小波神经网络的非线性拟合能力,构建了高效的语音识别模型。实证研究表明,该方案在TIMIT数据集上取得了80.17%的帧识别准确率,在满足实时性约束的同时,显著提升了嵌入式语音识别系统的性能表现,为智能语音交互在资源受限场景下的应用部署提供了新的思路。Aiming at the problem of resource limitation in embedded speech recognition,a lightweight recognition scheme based on wavelet neural network is proposed.This scheme uses wavelet transform to extract the time-frequency domain features of speech signals,and combines the nonlinear fitting ability of wavelet neural network to build an efficient speech recognition model.The empirical study shows that the scheme achieves 80.17%frame recognition accuracy on TIMIT data set,which not only meets the realtime constraints,but also significantly improves the performance of embedded speech recognition system,providing a new idea for the application deployment of intelligent voice interaction in resource-constrained scenarios.

关 键 词：嵌入式语音识别 小波变换 小波神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]