Comments on: Real FFT/IFFT with the Accelerate Framework

By: Valeriy Van

Valeriy Van — Wed, 10 Jul 2019 11:36:08 +0000

Call of `memset` before `vDSP_hann_window` is not needed.

By: jasvinder

jasvinder — Tue, 29 Mar 2016 09:34:34 +0000

nice and informative post. Thanks for such posts. Hope we will get more in future. Thanks Array In c language

By: Andrei

Andrei — Sun, 22 Sep 2013 10:57:48 +0000

Took a look at the hop size parameter. So setting it as -10 for example, causes it to overlap with the previous window by 10, right? And setting it as 10, causes it to skip 10…? Sorry for so many questions but i’m just starting with all this…

By: Parag K Mital

Parag K Mital — Thu, 19 Sep 2013 11:03:28 +0000

In reply to Andrei.

You can set the frame size when initializing the pkmFFT object.

As for the sample rate, this is up to your implementation. The FFT does not need to know this. You will get back magnitudes whose bins relate to your sample rate, but to calculate the magnitudes themselves, you do not need to know this.

For example, say your float* represents 8192 samples which sampled an analog signal at 10 KHz. Then we will have:

8192 / 2 = 4096 FFT bins
Via nyquist, our signal contains content up to 5khz and our bin resolution is:

5000Hz / 4096 bins = 1.22 Hz/bin

That is, each magnitude represents a width of 1.22 Hz…

Lastly, for frame overlap, this is only required when you calculate a short time fourier transform. See pkmSTFT and look for the hop size parameter.

Good luck…

By: Andrei

Andrei — Thu, 19 Sep 2013 09:59:31 +0000

Well…as I remember from school ( :D) a FFT has 3 parameters… Sample rate, frame width (in samples) and frame overlap. I want to set an overlapping of 95% (i know that this generates a lot of data, but I can always lower the sample rate). I just can’t find where I could do this.

By: Parag K Mital

Parag K Mital — Wed, 18 Sep 2013 21:27:43 +0000

In reply to Andrei. What exactly is the problem? You should send the frame of data to pkmFFT as a float*, and you will get back the magnitudes and phases as a float*.

By: Parag K Mital

Parag K Mital — Wed, 18 Sep 2013 21:26:52 +0000

In reply to Greg. Hi Greg, Sorry it should be back up now...

By: Andrei

Andrei — Wed, 18 Sep 2013 21:19:42 +0000

Hi. First of all, nice work. I’m still having some trouble getting this in my code. Here is my problem: I capture the sound using from the device’s microphone (audio units).I want to window each frame (Hann for example), and compute the FFT. I was thinking of using pkmFFT since it already contains Hann windowing, but I’m having trouble doing this. Think you might help? Thanks!

By: Greg

Greg — Fri, 16 Nov 2012 20:50:39 +0000

Hi, is this library no longer available? The github link returns a 404. 🙁

By: Parag K Mital

Parag K Mital — Wed, 22 Feb 2012 11:53:52 +0000

In reply to Chris. I believe the issue is with the way you've read the ".caf" audio file. Have a look at ExtAudioFile and AudioBufferList. These two should get you your float buffers fairly seamlessly. Hope that will help!