中圖分類號： TP399 文獻標識碼： A DOI：10.16157/j.issn.0258-7998.200960 中文引用格式： 牟俊杰，姚剛，孫濤. 基于CNN-LSTM神經(jīng)網(wǎng)絡的聲紋識別系統(tǒng)設計[J].電子技術應用，2021，47(3)：75-78. 英文引用格式： Mu Junjie，Yao Gang，Sun Tao. Design of vocieprint recognition system based on CNN-LSTM neural network[J]. Application of Electronic Technique，2021，47(3)：75-78.
Design of vocieprint recognition system based on CNN-LSTM neural network
Abstract： For warning of cardiovascular disease,in order to early detect the change of heart and lung voice representing the signs of danger,the vocieprint recognition system based on CNN-LSTM is designed. Using the Internet of Things technology coalescing the heart rate sensor chip, single-chip computer, electronic stethoscope, such as equipments,it can monitor the heart rate in real-time, early warn.And the cardiopulmonary sound recognition model based on the CNN-LSTM algorithm is trained, results show that the loss value is 0.082, accuracy rate of 0.908. The system is forward-looking and has a complete structural framework, which can effectively avoid the waste of medical resources, preposite the countermeasures for cardiovascular diseases.It has a broad application prospect in the market, and plays a significant role in promoting smart medical treatment.
Key words : CNN；LSTM；features extraction；MFCC；cardiovascular disease；vocieprint recognition
在人口老齡化日益嚴重的當下，心血管疾病不斷威脅老年人健康，引發(fā)社會廣泛關注。由于醫(yī)療知識欠缺、行動不便等原因，部分老年人就醫(yī)不及時，錯過了搶救的黃金時間，留下永遠的遺憾。開發(fā)心血管疾病方面的智能預警系統(tǒng)，滿足龐大的老年人群體需求迫在眉睫[3]。在醫(yī)療實踐中，對心血管疾病的診斷常常以心率、心肺音數(shù)據(jù)為重要支撐，國內(nèi)外以DSP[4]、長短時記憶(Long Short Time Memory，LSTM)[5]、卷積神經(jīng)網(wǎng)絡[6](Convolutional Neural Network，CNN)等方法算法為手段對心血管疾病的信號診斷進行了相當多的分析，但基本均停留在理論層面，距離軟硬件結合的實際應用尚有差距。各種醫(yī)療設備的聚焦點主要是信號的準確采集、分離[7-8]，基于醫(yī)療倫理等原因，對智能診斷設備的研制尚處于知識儲備期，有巨大的空白亟需填補。本文設計了基于CNN-LSTM的心血管疾病預警系統(tǒng)，利用物聯(lián)網(wǎng)技術采集心率和心肺音等健康指標數(shù)據(jù)，對老人的健康狀況進行實時監(jiān)測、預警，采用基于CNN-LSTM模型的智能算法對心肺音信號進行智能分析預警。系統(tǒng)著重考慮了適用性、穩(wěn)定性和成本，具有較高的實用價值和完整的結構框架，是利用智慧醫(yī)療從應用層面解決心血管疾病問題的一次重要探索。
