Question

我试图从音频文件中获取宽度为17的矩阵。我使用错误。*矩阵尺寸必须同意。它是什么意思，我想如何解决它？

错误位于windowed_frame = window.*input(((k-1)*bins*overlap)+1:(((k-1))*bins*overlap)+1+bins-1);

[input fs] = wavread('3_1.wav');

fn = fs/2;
bins = 512; %useable bins
frame_length = bins;
window = hamming(frame_length);
overlap = .5;
number_of_frames = (length(input)/(frame_length)/overlap) - 1;

for k=1:number_of_frames

        windowed_frame = window.*input(((k-1)*bins*overlap)+1:(((k-1))*bins*overlap)+1+bins-1);

        fft_window = fft(windowed_frame,bins*2);
        fft_window = abs(fft_window(1:bins));
        fft_energy = fft_window.^2;
        freq_vector = [fn/bins:fn/bins:fn];

编辑：完整代码 function test_energy = energy_per_frame（file）

[输入fs] = wavread（'1_2.wav'）;

fn = fs/2;
bins = 512; %useable bins
frame_length = bins;
window = hamming(frame_length);
overlap = .5;
number_of_frames = (length(input)/(frame_length)/overlap) - 1;

for k=1:number_of_frames

    windowed_frame = (window.*input(((k-1)*bins*overlap)+1:(((k-1))*bins*overlap)+1+bins-1)).';

        fft_window = fft(windowed_frame,bins*2);
        fft_window = abs(fft_window(1:bins));
        fft_energy = fft_window.^2;
        freq_vector = [fn/bins:fn/bins:fn];

        %map bark (z) to bins (1-512) and frequency (freq_vector)
        for i=1:length(freq_vector)
            z(i) = 13*atan(0.76*(freq_vector(i)/1000)) + 3.5*atan(((freq_vector(i)/1000)/7.5).^2);    
        end

        z = floor(z);

        for j=1:length(z);
            if z(j) == 17
                z(j)=16;
            end
        end

        temp = 0;
        temp_sum = 0;
        count = 0;

        %calculate the energy per bark band
        for i=1:length(z)
           if temp == z(i)
               temp_sum = fft_energy(i)+ temp_sum;
               test_energy(k,temp+1) = temp_sum; 
               count = count+1;
           end

           if temp ~= z(i)   
              test_energy(k,temp+1) = temp_sum/count;
              temp = temp+1;
              temp_sum = fft_energy(i);
              count = 1;
           end  
        end  

end

Answer 1

.*运算符是逐元素乘法。在您的情况下，这意味着window中的每个元素都乘以子矩阵input中的相应元素，这是操作的结果：

input(((k-1)*bins*overlap)+1:(((k-1))*bins*overlap)+1+bins-1)

显然，两个矩阵都有不同的维度，您可以通过输入以下内容来验证：

size(window)
size(input(((k-1)*bins*overlap)+1:(((k-1))*bins*overlap)+1+bins-1))

我怀疑这是一个行列问题：生成的子矩阵input是1 x 512行行向量，而windows是512-by- 1 列向量。

解决方案：尝试添加点撇号（.'），实际上是MATLAB中的transpose operator（与'运算符相对， complex-conjugate转置，如果你的数据很复杂，可能不是你想要的那样），在子矩阵input的末尾，如下所示：

windowed_frame = window .* input(blah blah...).'  %// <-- Note the .'

我相信这应该可以纠正你的问题。

更通用的解决方案是使用colon operator（:）强制两个向量成为列向量：

input_window = input(((k-1)*bins*overlap)+1:(((k-1))*bins*overlap)+1+bins-1);
windowed_frame = window(:) .* input_window(:);

使用时出错。*矩阵尺寸必须一致

1 个答案: