Matlab音频MFCC计算：函数调用与绘图实现指导请求

Got it, let's walk through building a complete main script to use your existing mfcc function. I'll cover loading audio, setting parameters, calling the function, and plotting the MFCC results clearly.

Step 1: Make Sure Dependencies Are Ready #

Your mfcc function calls two helper functions: vec2frames and trifbank. Double-check these are in your MATLAB path (same folder as your mfcc.m and main script) — if you don't have them, you can find standard implementations online or write simple versions, but I'll assume they're available for this walkthrough.

Step 2: Full Main Script Code #

Create a new .m file (e.g., run_mfcc.m) with this code:

% --------------------------
% Main Script to Compute and Plot MFCCs
% --------------------------

% 1. Load your audio file
% Replace 'your_audio.wav' with your actual audio file path
[speech, fs] = audioread('your_audio.wav');
% If audio is stereo, convert to mono
if size(speech, 2) > 1
    speech = mean(speech, 2);
end

% 2. Set MFCC parameters (match the function's input requirements)
Tw = 25;        % Frame duration in milliseconds
Ts = 10;        % Frame shift in milliseconds
alpha = 0.97;   % Pre-emphasis coefficient
window = @hamming; % Window function (hamming, hanning, etc.)
R = 0;          % Lowest frequency in Hz (usually 0)
M = 26;         % Number of mel filterbank channels
N = 13;         % Number of cepstral coefficients (including 0th)
L = 22;         % Cepstral lifter parameter

% 3. Call your MFCC function
[CC, FBE, frames] = mfcc(speech, fs, Tw, Ts, alpha, window, R, M, N, L);

% 4. Visualize the MFCC results
figure('Color','w');

% Plot MFCC spectrogram (transpose CC because columns are frames)
subplot(2,1,1);
imagesc(1:size(CC,2), 0:N-1, CC);
axis xy; % Flip y-axis to show 0th cepstrum at the top
colorbar;
title('MFCC Coefficients');
xlabel('Frame Number');
ylabel('Cepstral Coefficient Index');

% Plot Filterbank Energies (FBE) if you want to see intermediate results
subplot(2,1,2);
imagesc(1:size(FBE,2), 1:M, FBE);
axis xy;
colorbar;
title('Mel Filterbank Energies');
xlabel('Frame Number');
ylabel('Filterbank Channel Index');

% Optional: Plot a single frame's MFCC coefficients
figure('Color','w');
plot(0:N-1, CC(:,100)); % Plot 100th frame's MFCCs
title('MFCC Coefficients for Frame 100');
xlabel('Cepstral Coefficient Index');
ylabel('Amplitude');
grid on;

Step 3: What Each Part Does #

Let's break down the key bits:

• Audio Loading: audioread handles most common formats (WAV, MP3, etc.). We convert stereo to mono since MFCCs are typically computed on single-channel audio.
• Parameter Setup: I used standard values for speech processing, but you can tweak these:
  • Tw=25 and Ts=10 are industry standard for frame duration/shift.
  • alpha=0.97 is the default pre-emphasis coefficient to boost high frequencies.
  • M=26 filterbank channels and N=13 cepstral coefficients are typical for speech recognition.
• Function Call: We pass all required parameters to your mfcc function, which returns three outputs:
  • CC: The MFCC coefficients (size N x number_of_frames).
  • FBE: The mel filterbank energies (intermediate step).
  • frames: The windowed audio frames.
• Visualization:
  • imagesc is perfect for plotting MFCCs as a heatmap — transposing CC makes frames run along the x-axis, which is easier to interpret.
  • axis xy flips the y-axis so the 0th cepstral coefficient (which captures overall energy) is at the top.
  • The optional single-frame plot helps you inspect individual MFCC values.

Step 4: Testing and Troubleshooting #

• If you get an error about vec2frames or trifbank, make sure those functions are in your MATLAB path.
• If your audio is too quiet, you can normalize it before processing: speech = speech / max(abs(speech));
• Adjust parameters like M or N if you need more/less detail in your MFCCs.

内容的提问来源于stack exchange，提问作者yosra