What I'm trying to do is conceptually similar to the infamous NMIST classification example. Except that each digit is a computer generated sound wave.
I'll be adding some background noise to the data in order to improve real world accuracy.
My question is; considering the sequential data what model is best suited for this? Am I right in assuming a convolutional net would work?
I favour simpler model in exchange for a few percentage performance points and preferably it could be written with the Scikit Learn library.
Yes,
you can use 1D convolutional neural network. The convolutional filters can exploit consecutive parts of the signal and therefore it can be useful.
You can also try to look into recurrent neural networks which are more complex.