2
$\begingroup$

I'm making a handmade guitar chords dataset by recording 20480 samples of a 44.1kHz microphone to later on make a chord classification using CNN. But there's a problem... I'm making this dataset on my room. Quiet. When other noise enviroments comes to the case the classifier will give poor results. I want to apply some technique to reduce the noise, even if little. Do window functions reduce noise? Can someone help me by some advice of techniques/algorithms? Using Python, if possible...

$\endgroup$
1
$\begingroup$

It's probably better to train your classifier with recordings made in a very large variety of typical noisy situations, rather than to try to modify the input to match that with which the classifier was trained.

$\endgroup$
  • $\begingroup$ Often, for training clasification algorithms, to increase their generalization capabilities to unknown input data, it is useful to increase the amount of training data by adding perturbed versions of your original data, e.g., by adding different types of noise, reverberation, distortions/clipping. $\endgroup$ – applesoup Aug 26 '17 at 8:15
  • $\begingroup$ Does not the conv neural network already do it? $\endgroup$ – Denis Candido Oct 25 '17 at 9:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.