Spectrogram computed on the Bark scale

I want to generate a spectrogram for a given song, segmented at 1 second intervals, with 50% overlap.

I want to reproduce the results of a paper, which suggests computing the spectrogram on the 21 first Bark bands. My code is as follows, and I am not sure if it is correct. I would appreciate feedback on the process. I use 1:7700, so I can obtain a matrix with 7700 rows, which I can then segment into each individual bin. But I am not sure if the result returned makes sense.

[sounds,freqs]=audioread(file_name);
windowsize=floor(freqs); %if i want to change the windowsize to 0.5sec, I can divide by 2. So floor here is not required
%you can start from 0 or 20.
BandBarks = [20, 100, 200, 300, 400, 510, 630, 770, 920, 1080, 1270, 1480, 1720, 2000, 2320, 2700, 3150, 3700, 4400, 5300, 6400, 7700];

Spectro=spectrogram(sounds,hann(windowsize),floor(0.5*windowsize),1:7700);
%OR should I use
Spectro=spectrogram(sounds,hann(windowsize),floor(0.5*windowsize),1:7700,freqs);

Answer 1

[Nayan]

Hi

As I understand, you need to find the spectrogram on the bark-scale. This can be achieved by the following steps :-

1. Read the wave-file using audioread(filename)
2. Find the spectrogram of the signal with the desired "overlap", "number of frequency bins(nfft)" and the window type using spectrogram(x)
3. Window length and the overlap can be calculated as per the requirement.
4. Once the spectrogram is obtain, the spectrogram on bark-scale can be obtained using hz2bark(hz)

I would suggest you to take help of the following code snippet and the libraries mentioned above :-

N = 1024;
n = 0:N-1;
[x, fs] = audioread('farspeech.wav');
duration = length(x)/fs;
t = linspace(0, duration, length(x));
plot(t, x);
xlabel('Time (s)');
ylabel('Amplitude');
window = hann(256);
noverlap = 128; % can be adjusted as needed
nfft = 512; % can be adjusted as needed
spectrogram(x, window, noverlap, nfft, fs, 'yaxis');
[s, f, t] = spectrogram(x, window, noverlap, nfft, fs, 'yaxis');
f_bark = hz2bark(f);
imagesc(t, f_bark, 20*log10(abs(s)));
axis xy;
xlabel('Time (s)');
ylabel('Bark');
colorbar;

You can also obtain spectrogram with different scales directly by using simulink block. I would suggest you to go through the following link for you benifit and interest.

Extract mel, Bark, or ERB spectrogram from audio - Simulink (mathworks.com)

Hope this helps!