✅作者简介：热爱科研的Matlab仿真开发者，修心和技术同步精进，代码获取、论文复现及科研仿真合作可私信。
🍎个人主页：Matlab科研工作室
🍊个人信条：格物致知。
🔥 内容介绍
声音自动分类是语音识别领域的一项重要任务，广泛应用于语音交互、语音控制和医疗诊断等领域。本文提出了一种基于主成分分析（PCA）和最近邻（KNN）相结合的声音自动分类方法。PCA用于提取声音特征的降维表示，KNN用于基于降维特征进行分类。实验结果表明，该方法在多个声音数据集上取得了较高的分类精度，证明了其有效性和实用性。
引言
声音自动分类旨在根据声音特征将声音样本分类到预定义的类别中。传统的声音自动分类方法通常依赖于手工提取的特征，这需要大量的专业知识和经验。近年来，随着机器学习和深度学习的发展，基于数据驱动的特征提取和分类方法得到了广泛的关注。
方法
本文提出的方法包括以下步骤：
- **特征提取：**使用梅尔频谱系数（MFCC）从声音样本中提取特征。MFCC是一种广泛用于语音识别领域的特征提取算法，它可以有效地捕捉声音的频谱信息。
- **主成分分析（PCA）：**对MFCC特征进行PCA降维。PCA是一种线性变换，它可以将高维特征投影到低维子空间中，同时保留最大方差的信息。PCA降维可以减少特征的冗余和噪声，提高分类的效率和准确性。
- **最近邻（KNN）：**使用KNN算法对PCA降维后的特征进行分类。KNN是一种非参数分类算法，它将一个新的样本分类为与它最相似的K个样本所属的类别。
结论
本文提出了一种基于PCA和KNN相结合的声音自动分类方法。该方法利用PCA降维提取声音特征的有效表示，并使用KNN进行分类。实验结果表明，该方法在多个声音数据集上取得了较高的分类精度，证明了其有效性和实用性。该方法可以为语音识别、语音控制和医疗诊断等领域提供一种有价值的工具。
📣 部分代码
clear; clc
% ––––––––––––––––––––––––––––– IASPROJECT –––––––––––––––––––––––––––––––
%----AUTHOR: ALESSANDRO-SCALAMBRINO-923216-------
% ---MANAGE PATH, DIRECTORIES, FILES--
addpath(genpath(pwd))
fprintf('Extracting features from the audio files...\n\n')
coughingFile = dir([pwd,'/Coughing/*.ogg']);
cryingFile = dir([pwd,'/Crying/*.ogg']);
snoringFile = dir([pwd,'/Snoring/*.ogg']);
F = [coughingFile; cryingFile; snoringFile];
% ---WINDOWS/STEP LENGHT---
windowLength = 0.025;
stepLength = 0.01;
% ---FEATURES EXTRACTION---
% ---INITIALIZING FEATURES VECTORS---
allFeatures = [];
coughingFeatures = [];
cryingFeatures = [];
snoringFeatures = [];
% ---EXTRACTION---
for i=1:3
for j=1:40
Features = stFeatureExtraction(F(i+j-1).name, windowLength, stepLength);
allFeatures = [allFeatures Features];
if i == 1; coughingFeatures = [coughingFeatures Features]; end
if i == 2; cryingFeatures = [cryingFeatures Features]; end
if i == 3; snoringFeatures = [snoringFeatures Features]; end
end
end
% ---FEATURES NORMALIZATION----
mn = mean(allFeatures);
st = std(allFeatures);
allFeaturesNorm = (allFeatures - repmat(mn,size(allFeatures,1),1))./repmat(st,size(allFeatures,1),1);
% ---PCA---
warning('off', 'stats:pca:ColRankDefX')
[coeff,score,latent,tsquared,explained] = pca(allFeaturesNorm');
disp('The following results are the values of the variance of each coefficient:')
explained
counter = 0;
for p=1:length(explained)
if explained(p) > 80
counter = counter + 1;
end
end
disp(['The number of coefficients offering at least 80% of variance is ', mat2str(counter)])
fprintf('\n\n')
% ---PCA PLOTTING---
S=[]; % size of each point, empty for all equal
C=[repmat([1 0 0],length(coughingFeatures),1); repmat([0 1 0],length(cryingFeatures),1); repmat([0 0 1],length(snoringFeatures),1)];
scatter3(score(:,1),score(:,2),score(:,3),S,C,'filled')
axis equal
title('PCA')
% –––––––––––––––––––––––TRAIN/TEST DATASET –––––––––––––––––––––––
trainPerc = 0.70;
testPerc = 1 - trainPerc;
coughingTrain = coughingFile(1:length(coughingFile)*trainPerc);
cryingTrain = cryingFile(1:length(cryingFile)*trainPerc);
snoringTrain = snoringFile(1:length(snoringFile)*trainPerc);
FTR = [coughingTrain cryingTrain snoringTrain];
TEST DATASET NORMALISATION –––––––––––––––––––––
% normalisation in time domain of TEST data
allTestTimeFeatures = allTestTimeFeatures';
allTestTimeFeatures = (allTestTimeFeatures - repmat(mnTime,size(allTestTimeFeatures,1),1))./repmat(stTime,size(allTestTimeFeatures,1),1);
% normalisation in frequency domain of TEST data
allTestFreqFeatures = allTestFreqFeatures';
allTestFreqFeatures = (allTestFreqFeatures - repmat(mnFreq,size(allTestFreqFeatures,1),1))./repmat(stFreq,size(allTestFreqFeatures,1),1);
% normalisation of both time domain and frequency domain of TEST data
allTestFeatures = allTestFeatures';
allTestFeatures = (allTestFeatures - repmat(mnAll,size(allTestFeatures,1),1))./repmat(stAll,size(allTestFeatures,1),1);
% –––––––––––––––––––––––––– TRAIN/TEST LABELS ––––––––––––––––––––––––––
% TRAIN
labelcoughingTime = repmat(1,length(coughingTrainTimeFeatures),1);
labelcryingTime = repmat(2,length(cryingTrainTimeFeatures),1);
labelsnoringTime = repmat(3, length(snoringTrainTimeFeatures),1);
allTimeLabels = [labelcoughingTime; labelcryingTime; labelsnoringTime];
labelcoughingFreq = repmat(1,length(coughingTrainFreqFeatures),1);
labelcryingFreq = repmat(2,length(cryingTrainFreqFeatures),1);
labelsnoringFreq = repmat(3, length(snoringTrainFreqFeatures),1);
allFreqLabels = [labelcoughingFreq; labelcryingFreq; labelsnoringFreq];
labelcoughingAll = repmat(1,length(coughingTrainFeatures),1);
labelcryingAll = repmat(2,length(cryingTrainFeatures),1);
labelsnoringAll = repmat(3, length(snoringTrainFeatures),1);
allLabels = [labelcoughingAll; labelcryingAll; labelsnoringAll];
% ––––––––––––––––––––––––––– APPLY TEST LABELS ––––––––––––––––––––––––––
testLabelcoughingTime = repmat(1,length(coughingTestTimeFeatures),1);
testLabelcryingTime = repmat(2,length(cryingTestTimeFeatures),1);
testLabelsnoringTime = repmat(3, length(snoringTestTimeFeatures),1);
groundTruthTime = [testLabelcoughingTime; testLabelcryingTime; testLabelsnoringTime];
testLabelcoughingFreq = repmat(1,length(coughingTestFreqFeatures),1);
testLabelcryingFreq = repmat(2,length(cryingTestFreqFeatures),1);
testLabelsnoringFreq = repmat(3, length(snoringTestFreqFeatures),1);
groundTruthFreq = [testLabelcoughingFreq; testLabelcryingFreq; testLabelsnoringFreq];
testLabelcoughingAll = repmat(1,length(coughingTestFeatures),1);
testLabelcryingAll = repmat(2,length(cryingTestFeatures),1);
testLabelsnoringAll = repmat(3, length(snoringTestFeatures),1);
allGroundTruth = [testLabelcoughingAll; testLabelcryingAll; testLabelsnoringAll];
% –––––––––––––––––––––––––––– KNN –––––––––––––––––––––––––––
fprintf('––––––––––––––––––––––––––– COMPUTING THE KNN ––––––––––––––––––––––––––––\n\n')
fprintf('Computing the recognition rate using the following values for k: 1, 2, 3, 5, 7, 8, 10, 15, 20, 50, 100, 200...\n\n')
%TIME
KNN_calculation(allTrainTimeFeatures, allTestTimeFeatures, allTimeLabels, groundTruthTime, testLabelcoughingTime, testLabelcryingTime, testLabelsnoringTime, 'TIME DOMAIN', '-pm')
%FREQ
KNN_calculation(allTrainFreqFeatures, allTestFreqFeatures, allFreqLabels, groundTruthFreq, testLabelcoughingFreq, testLabelcryingFreq, testLabelsnoringFreq, 'FREQUENCY DOMAIN', '-pg')
%ALL-TOGETHER
KNN_calculation(allTrainFeatures, allTestFeatures, allLabels, allGroundTruth, testLabelcoughingAll, testLabelcryingAll, testLabelsnoringAll, 'TIME AND FREQUENCY DOMAIN', '-pr')⛳️ 运行结果
🔗 参考文献
[1]王心醉.人脸识别算法在ATM上的应用研究[J]. 2009.DOI:http://159.226.165.120//handle/181722/1057.
🎈 部分理论引用网络文献，若有侵权联系博主删除
🎁 关注我领取海量matlab电子书和数学建模资料
1 各类智能优化算法改进及应用
生产调度、经济调度、装配线调度、充电优化、车间调度、发车优化、水库调度、三维装箱、物流选址、货位优化、公交排班优化、充电桩布局优化、车间布局优化、集装箱船配载优化、水泵组合优化、解医疗资源分配优化、设施布局优化、可视域基站和无人机选址优化、背包问题、 风电场布局、时隙分配优化、 最佳分布式发电单元分配、多阶段管道维修、 工厂-中心-需求点三级选址问题、 应急生活物质配送中心选址、 基站选址、 道路灯柱布置、 枢纽节点部署、 输电线路台风监测装置、 集装箱船配载优化、 机组优化、 投资优化组合、云服务器组合优化、 天线线性阵列分布优化
2 机器学习和深度学习方面
2.1 bp时序、回归预测和分类
2.2 ENS声神经网络时序、回归预测和分类
2.3 SVM/CNN-SVM/LSSVM/RVM支持向量机系列时序、回归预测和分类
2.4 CNN/TCN卷积神经网络系列时序、回归预测和分类
2.5 ELM/KELM/RELM/DELM极限学习机系列时序、回归预测和分类
2.6 GRU/Bi-GRU/CNN-GRU/CNN-BiGRU门控神经网络时序、回归预测和分类
2.7 ELMAN递归神经网络时序、回归\预测和分类
2.8 LSTM/BiLSTM/CNN-LSTM/CNN-BiLSTM/长短记忆神经网络系列时序、回归预测和分类
2.9 RBF径向基神经网络时序、回归预测和分类
