I am trying to implement speech recognition but I'm struck with a few questions:

  1. There are 12 Coefficients of MFCC. What are their names?
  2. What is the range of values of MFCC coefficients?
  3. If I want to relate MFCC coefficients graphically, what will be unit of weights on their edges?
  • $\begingroup$ If question 3 refers to the axes of a graph: x-axis is quefrency, y-axis is amplitude on a log scale. $\endgroup$ Jul 6, 2014 at 9:32

1 Answer 1

  1. You can have as many MFCC coefficients as you want, 12 is just only a widely used number. As you probably know (or if not then please refer to the old answer), coefficients are being obtained via fitting of cosinusoids to log energies in your filter banks. Usually 12 is enough, more is not improving recognition rate too much. It is better to calculate their derivatives ($\Delta $'s and $\Delta\Delta$'s across time). Regarding names - they are simply MFCC's (first, second, and so on). Although coefficient $c_0$ is usually called the energy.
  2. Question about range is not straightforward to answer. Usually in my applications they are in $[-12, \ 12] $. Although if you perform the Cepstral Mean Normalisation on yours, then you can expect them to have values of Gaussian Distribution with $95\%$ of them being in range of $[-2, \ 2] $.
  3. I do not really understand that question - you must be more specific about what you are asking. Edges of what?

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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