1
$\begingroup$

Task:

Multi-label classification of sounds using neural networks. (Urbansound8K Dataset)

Problem:

How to best generate my combined dataset, considering maximum 2 sounds combined at the same time.

Say I have 9 sounds belonging to 3 different classes (A,B,C):

A1 A2 A3

B1 B2 B3

C1 C2 C3

Options:

1) Not repeat any file. This would give 4 combinations (8 files) and 1 leftover file. For example: A1B1, A2B2, A3C2, B3C3. Leftover C1.

2) Repeat files, but only with one other example of each different class. This would give 9 combinations: A1B1, A1C1, A2B2, A2C2, A3B3, A3C3, B1C1, B2C2, B3C3.

3) Repeat files as many times as there are files of other classes. For example, combining the example A1 would give: A1B1, A1B2, A1B3, A1C1, A1C2, A1C3. I believe this option would give in total 27 examples.

Questions:

1) Which option would the the optimal for training a neural network for multi-label classification of the classes A,B and C?

2) Is there any library that would make this automatically?

Thanks for any help!

$\endgroup$
  • $\begingroup$ what are these files and combinations you refer to? $\endgroup$ – shimao May 17 '18 at 20:17
  • $\begingroup$ Sorry yeah, I should have mentioned that. These are audio signals @shimao $\endgroup$ – sdiabr May 18 '18 at 6:56

Your Answer

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

Browse other questions tagged or ask your own question.