基于深度学习的LED车载语音交互氛围灯设计

A Design of Voice Interactive LED Automotive Ambient Lighting Based on Deep Learning

由于常用车载氛围灯的操作方式局限于触摸和按键操作,采用语音交互模式的智能车载氛围灯开始受到关注。本文采用深度学习方法设计了一种LED车载语音交互氛围灯,并在嵌入式平台实现了控制识别与控制原理。首先,采用深度学习库keras和卷积神经网络(convolutional neural network,CNN)对离线语音进行训练得到声学模型,并移植该模型至嵌入式平台中;其次,对采集语音信号进行数据预处理,采用梅尔频率倒谱系数(mel frequency cepstral coefficient,MFCC)方法进行特征提取,并利用声学模型对提取特征进行识别分类;最后,通过CAN总线获取汽车故障、车速等车身信息,并将各类信息综合判断后实现车载氛围灯的各类灯光效果控制。

Since the operation of common automotive ambient lighting is limited to the touch and button modes,the voice interactive ambient lighting has attracted more and more attentions.In this paper,a deep learning method is adopted to design a voice interactive LED automotive ambient lighting,and the control recognition and control principle are realized on an embedded platform.Firstly,the offline speech was trained by using the keras and Convolutional Neural Network(CNN)to obtain the acoustic model,which was then transplanted into the embedded platform.Secondly,the collected voice signals were preprocessed.The features were extracted by using the Mel Frequency Cepstral Coefficent(MFCC)method,and the extracted features were identified and classified through the acoustic model.Finally,vehicle fault,speed and other information were obtained through the CAN bus,and various lighting controls can be achieved after a comprehensive judgment of all kinds of information.