Re: the autocorrelation function for measuring the F0

["Richard F. Lyon" <DickLyon@xxxxxxx>]

Title: Re: the autocorrelation function for measuring the F0

At 1:10 PM -0700 7/28/05, Stefan Crawcour wrote:

> When calculating the F0 from the speech
> sample, I automatically use the autocorrelation function. But, what
> does the autocorrelation function really do? Is it a similar mechanism
> as in the FFT (multiplying the speech sample with an appropiate
> window, if I am not terribly mistaken) ?

It's not clear what you mean when you say you "use the autocorrelation function."  Speech does not, in general, have an autocorrelation function, which is defined as a function of a stationary random process, or of a finite signal (with bounded energy); speech is neither.  What you're probably doing is a short-time autocorrelation, which is defined for signals of finite duration, or for segments of signals made finite by windowing.

Here's some stuff google finds for short-time-autocorrelation-function:

http://shay.ecn.purdue.edu/~ee649/homework/hw3.pdf

http://home.netnam.vn/elib/index.asp?progid=305&page=41&id=40876

http://www.nis.sdu.dk/~marcela/NIS04/week-40/lesson_4.pdf

It is not unusual for the phrase "short-time" to be omitted, thereby confusing the issue.

To answer "what does it really do?": it computes of the ACF of the windowed signal, which might be what you were wanting to verify.  Or was it?

Dick