0
$\begingroup$

i am working on a project related to audio events recognition in real time like a door bell, baby crying, footstep, I have 9 categories of sound events so the first step i did was getting many wav files for the categories i want to recognize (the length of each audio file is 2-3 seconds),then i ran the MFCC function for every single audio file and took the average of it and added it to an excel sheet.

the second step was making a test for that model using the Gaussian methods(by matlab)

here i am making a test by 90%training files and 10% testing files

here i am making a test by 90%training files and 10% testing files

here i am making a test by 80%training files and 20% testing files enter image description here

so, why am I getting low recognition results ,and is there anyway to improve it? thanks

| improve this question | |
$\endgroup$
  • $\begingroup$ Averaging of MFCC’s does not make much sense. Train the GMM’s on all frames which belong to each class and then perform frame wise scoring. $\endgroup$ – jojek♦ Jul 8 '18 at 15:31
  • $\begingroup$ thanks sir, i will do it but what do you mean by "perform frame wise scoring" $\endgroup$ – KamelK Jul 8 '18 at 17:38
  • $\begingroup$ Calculate the log likelihood for all input frames given GMM $\endgroup$ – jojek♦ Jul 8 '18 at 17:43

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.