I am using HTK to recognize which word has been said.
I am working on 20 words and I have for each of them 10 .wav files.
Here are the steps that i followed: 1) get the MFC file for each .wav file for each word 2) for each word train it on the 10 MFCs files 3) add the grammar and dictionary and hmmlist in the HTK 4) test
in the test, i use the mfc of the training data, but i get an error of 3.5% when i use a new mfc (not in the training data) it never recognized it right :S
Any ideas of what to do ?