# Obtaining signal energy modulation for voice activity detection

Im currently working on VAD (voice activity detection) topic. Since I'm beginner in digital signal processing, I've started to wonder if calculating single frame energy (after aplying FFT to it) and looking for values around 4Hz and bandwith (for example) from 200Hz to 600Hz would be as effective as checking for 4Hz modulation energy after finding MFCC coefficients?

## 1 Answer

From my understanding of your question, you want to ask whether one should look for energy(of FFT coefficients) of bands directly, or one should use MFCC's (energy) for making VAD decisions.

As MFCC based approach is to mimic the human auditory systems response, it will be better to use MFCC coefficients.

if one uses same bands as MEL Filterbank for calculating band-wise energy (after taking FFT), than the results will be similar to using MFCC'S, as MFCC's are just the transformation from FFT coefficients.

• Thanks for answer. So - is it even necessary to calculate of energy of FFT coefficients or should I just pass them to MFCC algorithm? Dec 31, 2015 at 12:34
• I think there is some confusion, you can either use FFT coefficients(after passing through MEL filter bank) for energy calculation, or you can use MFCC Coefficients. So, if you want to use MFCC's no need to calculate energy of FFT coefficients. Dec 31, 2015 at 12:52
• Yes, I meant that better solution would be finding FFT coefficients and then feed with them MFCC algorith for obtaining MFCC coefficients (as you proposed), then calculating its energy in 4Hz frequency, right? Dec 31, 2015 at 13:03
• I actually don't know which algorithm you are using for finding MFCC coefficients, if your algorithm takes FFT coefficients as input and gives MFCC's as output, go ahead with it. Regarding energy calculation in 4Hz, I don't know what literature you are referring to for making VAD decisions, if possible provide the link for your reference. Dec 31, 2015 at 13:14
• I think I'm going to use this code for computing MFCC coefficients - code.google.com/p/libmfcc/source/browse/libmfcc.c And for VAD decisions, I'm going to combine zero-crossing ratio, entropy modulation and the 4Hz modulation energy I meant. Dec 31, 2015 at 14:00