I was making a simple phoneme classification model for a 10 week-long class project and I ran into a small question.

Is it possible to create a model that takes a 1-second (the longest phoneme is 0.2 second but the large image is kept for context) spectrogram as input? Some people suggest creating an RNN for phoneme classification, but can you build a pure CNN phoneme classification model?


1 Answer 1


Yes you can, a few years ago I made a simple CNN for a single Arabic phoneme classification. You can use spectogram or using MFCC / MFSC as features, as long all data has the same size (use padding or cropping if needed).

You may need RNN if you want to combine some phonemes to recognize a single word or longer.

  • $\begingroup$ Wow, thanks a lot! Any suggestions for the model shape (how many layers and how many filters)? $\endgroup$ Nov 17, 2019 at 3:09
  • $\begingroup$ Actually, it all depends on your case, for a simple case I think you only need one or two convolutional layers and a single fully-connected layer. I usually use around 30 filters. But, just try it to find your best values :) $\endgroup$
    – malioboro
    Nov 17, 2019 at 3:16
  • $\begingroup$ Okay, well I actually I have a full model written down but it isn't running well (not increasing in accuracy). Where should I post my code and give a better description of my model? $\endgroup$ Nov 17, 2019 at 3:52
  • $\begingroup$ You can ask another question if you think there are some errors in your models. But, you may ask an implementation question that out of topic in AI.SE if your question is about errors in your code/implementation, you should ask that in Data Science SE $\endgroup$
    – malioboro
    Nov 17, 2019 at 7:39
  • $\begingroup$ Okay thanks, I put the question on this link, so you can look over it if you want to, 'datascience.stackexchange.com/questions/63376/…' $\endgroup$ Nov 19, 2019 at 3:32

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .