This question already has an answer here:

I am doing the project on speech recognition.I have implemented both the mfcc features and neural network.I am a bit confused in joining the two blocks.For example i have three different words with their '.wav' files.I extract 13 mfcc features from these three words respectively.There will be a matrix of mfcc e.g (13x100) for each word.I convert this matrix to a vetor.Than i will have three vectors of 1300 length (assuming each word is of same length).Than i pass these vectors as input to my neural network.I am confused about my training output.Whether it will be the same wav files from which i extracted the mfcc features or will it be something else.


marked as duplicate by Nikolay Shmyrev, Matt L., Peter K. Dec 5 '15 at 23:15

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ Usually the neural network is trained to recognize smaller units rather than a whole word. $\endgroup$ – Aaron Aug 11 '14 at 20:14
  • $\begingroup$ Read a current tutorial. The "smaller units" @Aaron refers to are called phonemes. $\endgroup$ – Emre Nov 9 '14 at 20:15

Usually your training data and your test data are different, otherwise your results will be unrealistically good. However, for a first test of your implementation, you can simply use your training data as test material. In this case, the recognition performance should be very good. If this is not the case you know that something went wrong.


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