1
$\begingroup$

I have just a quick question about how to calculate Deltas from MFCC. I have found the formula for calculating deltas here:

$$d_t=\frac{\sum_{n=1}^{N}n\left(c_{t+n}-c_{t-n}\right)}{2\sum_{n=1}^{N}n^2}$$

Another source states it is a regression equation. From what I can see, it takes coefficient from a previous and next frame. For any $N$ e.g. $N=2$, $t-n$ in $c_{t-n}$ can refer to the past where there are no frames. Which values should I substitute in this case? Should it be simply 0?

Similarly, when $t+n$ in $c_{t+n}$ refers to a frame in future that not exists, should it also be 0?

Thanks

Improve this question
$\endgroup$
4
$\begingroup$

It's better to copy first frame and last frame values to extend vector sequence beyond boundaries than to assign 0. This could be implemented just by adjusting indexes:

if (index1 < 0)
    index1 = 0
if (index2 > N - 1)
   index2 = N - 1

delta = v[index1] - v[index2]
Improve this answer
$\endgroup$
1
  • $\begingroup$ Yes. 0 are not typical values for MFCC, so using 0 for the first/last frames would give spurious values of the delta value. $\endgroup$ – pichenettes Jan 24 '14 at 13:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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