I was reading this paper, and in section 3 I got a bit confused. It claims that the algorithmic latency is equal to synthesis frame length. If the overlap between two consecutive frames is 50%, the algorithmic latency should be the half of the window size. What am I misunderstanding here?

  • $\begingroup$ it is not about the latency of STFT. $\endgroup$ – jojek Sep 5 '17 at 20:38
  • $\begingroup$ @jojek can you explain a bit more? $\endgroup$ – Sounak Sep 5 '17 at 20:46
  • $\begingroup$ The delay is independent of the overlap because you have to wait for a complete block to be filled with new input data before you can do the [transformation$\rightarrow$processing$\rightarrow$inverse transformation] and output a new block of output data which consists - in the case of overlap-add - of the sum of the first samples of the current block and the last samples of the previous last block. (Compare this figure from the Wikipedia page on the overlap-add algorithm.) $\endgroup$ – applesoup Sep 20 '17 at 17:46

Your Answer

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

Browse other questions tagged or ask your own question.