Classifying environmental sounds using AlexNet

I am trying to extract Spectrograms and MFCC from audio files then calssify them using AlexNet or GooglNet. Any tips ?

Answer 1

[Utkarsh]

Hi Ibrahim,

For classification of spectrogram with alexnet, you might wish to train the last layers of alexnet according to your dataset (spectrograms of audio). For more details about transfer learning with alexnet refer here.

You may also refer here for an approach without alexnet.

[Ibrahim A]

Hi Utkarsh,

Thank you for your response. But I am not sure how to convert my audio dataset from .wav to images so I can classify them with AlexNet. Check out the attached paper to see what I exactly mean.

Thanks

[Utkarsh]

You may refer here for documentation on how to create a spectrogram for an audio

Have you tried this for saving the spectrogram? (It will save the spectrogram without any border,label,etc)

s = spectrogram(x); %x contains the loaded audio in an array
spectrogram(x,'yaxis')
set(gca, 'Visible', 'off');
colorbar('off');
saveas(gcf,'file_name.png') 

[Ibrahim A]

Thank you Utkarsh