Speech/Music Discrimination

Hello,

I have a general problem on discriminating speech and music signal on matlab. I need to create FFT/frequency graphs for each music and speech files and then create Spectral Centrum graph which must show average frequency difference between each that is how i discriminate them. I would appreciate if you help me.

[Thiago Henrique Gomes Lobato]

You need to be more specific in your question, what have you tried? How exactly you want to discriminate them, just looking at a graph? Sound discriminating is a very complex problem so you need to be more specific. If all that you want are the frequency bins take a look at the Spectogram function

[Ömer Kaan Karaalp]

I have 4 audio files which are 2 human voices and 2 musics. Firstly, I want to create 4 fft/frequency graph and then by using spectral centroid, i want to find out average frequencies. As a result, there must be 4 averages but there must also be specific interval between music and human voice signals. That is how i want to discriminate them but i do not know how to implement these on matlab. thanks for your attention sir.

Answer 1

[Brian Hemmat]

As a start, take a look at the spectralCentroid function in Audio Toolbox™.

This tutorial also covers some aspects of the area you're looking into (it has an example that uses spectralEntropy for music/speech discrimination):

https://www.mathworks.com/help/audio/ug/spectral-descriptors.html#SpectralDescriptorsExample-5

Below is some simple code to get you started. If you look at the histogram, you can see that a good threshold for the spectral centroid, based on this limited data, is around 330 Hz. That is, if the average spectral centroid for some audio is above 330 Hz, declare it as music. If it is below, declare it as speech. The code below requires Audio Toolbox™ R2019a or above to run.

desiredFs = 16e3;

[music1,fs] = audioread('RockGuitar-16-44p1-stereo-72secs.wav');
music1 = resample(mean(music1,2),desiredFs,fs);

[music2,fs] = audioread('handel.ogg');
music2 = resample(music2,desiredFs,fs);

[speech1,fs] = audioread('Rainbow-16-8-mono-114secs.wav');
speech1 = resample(speech1,desiredFs,fs);

[speech2,fs] = audioread('SpeechDFT-16-8-mono-5secs.wav');
speech2 = resample(speech2,desiredFs,fs);

music1_centroid = spectralCentroid(music1,fs);
music2_centroid = spectralCentroid(music2,fs);
speech1_centroid = spectralCentroid(speech1,fs);
speech2_centroid = spectralCentroid(speech2,fs);

music_centroid = [music1_centroid;music2_centroid];
speech_centroid = [speech1_centroid;speech2_centroid];

figure
h1 = histogram(music_centroid,'Normalization','probability');
hold on
h2 = histogram(speech_centroid,'Normalization','probability');
legend('Music','Speech')
xlabel('Centroid (Hz)')
ylabel('Probability')