3
$\begingroup$

I'm working in speech recognition research, and I wonder if there is a way to analyse sound just like how ears do (i.e. without windowing), so for example to take some feature continuously with an update for each new sample.

It is similar to talk about windowing with overlapping ratio of 99% or something like this, you can look at the question like: How does ear working without windowing concept?

$\endgroup$
  • $\begingroup$ wavelets might be fine $\endgroup$ – LJSilver May 6 '16 at 12:18
  • $\begingroup$ can you provide a link for this specific feature (window-less transform) ? $\endgroup$ – Mostafa 36a2 May 6 '16 at 12:23
7
$\begingroup$

Yes, it's possible to analyse sound the way ears do.

For example, you could compute the DFT of a signal continuously using several Goertzel filters.

$$ y_k[n] = e^{j2\pi k/N} y_k[n-1] + x[n] $$

where $k= 0,1,\ldots, N-1$, so that $y_0$ is the DC or zero frequency term.

Of course, this is an unstable filter, so some resetting or forgetting factor is needed to keep it stable. There are ways to make it better behaved numerically (see pp84-93 of Stoer & Bulirsch).

The reason why we don't generally do things this way is that we are usually trying to do them in the most computationally efficient way possible. And nature doesn't necessarily do this, so we find ways to do similar things but that work better on a computer.

$\endgroup$

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.