语音识别代码实现Matlab

以下是一个基于 MATLAB 的语音识别示例代码，使用了 MFCC 特征提取和 GMM-HMM 模型来进行分类。

% 读取训练数据和测试数据
train_files = dir('train/*.wav');
test_files = dir('test/*.wav');

% 特征提取参数
frame_length = 25; % 每帧的长度（ms）
frame_shift = 10; % 帧之间的相邻距离（ms）
num_mfcc = 12; % MFCC 系数的数量

% 特征提取
train_features = [];
train_labels = [];
for i = 1:length(train_files)
    filename = fullfile(train_files(i).folder, train_files(i).name);
    [audio, fs] = audioread(filename);
    mfccs = mfcc(audio, fs, 'WindowLength', round(fs * frame_length / 1000), ...
        'OverlapLength', round(fs * (frame_length - frame_shift) / 1000), ...
        'NumCoeffs', num_mfcc, 'LogEnergy', 'Ignore');
    train_features = [train_features; mfccs];
    label = strsplit(train_files(i).name, '_');
    train_labels = [train_labels; label{1}];
end

% 模型训练
num_states = 5; % 隐状态数
num_mixtures = 3; % 混合高斯数
gmm = fitgmdist(train_features, num_mixtures, 'CovarianceType', 'diagonal', ...
    'RegularizationValue', 0.1, 'Options', statset('MaxIter', 100));
hmm = fitHMM(train_features', num_states, 'gaussian', 'mixtures', num_mixtures, ...
    'SharedCov', false, 'Verbose', false, 'maxiter', 100);

% 测试数据分类
test_labels = [];
for i = 1:length(test_files)
    filename = fullfile(test_files(i).folder, test_files(i).name);
    [audio, fs] = audioread(filename);
    mfccs = mfcc(audio, fs, 'WindowLength', round(fs * frame_length / 1000), ...
        'OverlapLength', round(fs * (frame_length - frame_shift) / 1000), ...
        'NumCoeffs', num_mfcc, 'LogEnergy', 'Ignore');
    log_likelihoods = log(pdf(gmm, mfccs))';
    decoded_sequence = viterbi(hmm, log_likelihoods);
    test_labels = [test_labels; decoded_sequence{1}];
end

% 输出分类结果
for i = 1:length(test_files)
    fprintf('%s is classified as %s\n', test_files(i).name, test_labels{i});
end

注意：该示例代码需要用到 Signal Processing Toolbox 和 Hidden Markov Model Toolbox。如果没有安装，请先进行安装。