Eldar Sadykov

I understand that pfft~ object receives a collection of time-domain samples and transforms them into a frequency-domain data and then doing the inverse of that. I've interacted a lot with 

patchers and the resynthesis happens frame-by-frame there derived from the matrix data. And when it happens, either time or frequency details loss happens. I just wonder why doesn't that happen when we perform for example fft filtering like in Forbidden planet example? I clearly hear that all the details of the original sound remain there except it becomes filtered. Even if I use really small fft sizes, if I just pass the audio through the pfft~ without any processing, it just delivers exactly the same sound and no time smearing or frequency information loss happens. It seems like I don't get something really basic about fft.

I understand that pfft~ object receives a collection of time-domain samples and transforms them into a frequency-domain data and then doing the inverse of that. I've interacted a lot with Jean-Francois Charles tutorial patchers and the resynthesis happens frame-by-frame there derived from the matrix data. And when it happens, either time or frequency details loss happens. I just wonder why doesn't that happen when we perform for example fft filtering like in Forbidden planet example? I clearly hear that all the details of the original sound remain there except it becomes filtered. Even if I use really small fft sizes, if I just pass the audio through the pfft~ without any processing, it just delivers exactly the same sound and no time smearing or frequency information loss happens. It seems like I don't get something really basic about fft.

I understand that pfft~ object receives a collection of time-domain samples and transforms them into a frequency-domain data and then doing 

basic-fft-question

That's the magic of the digital Fourier math: even with an extremely small FFT size (say, 16 or even 2), if you don't do any processing, the re-synthesis is exactly the same as the original signal! The information in frequency domain is the same as the information in the source signal, and as long as you don't change it, all of this information is back to the time domain after an inverse FFT. Now, if you do process the sound, things are different.

The Forbidden planet example is very interesting, especially if you pair it with Mark Dolson's Computer Music Journal article "The Phase Vocoder: a Tutorial". Just a bunch of bandpass filters.

That's the magic of the digital Fourier math: even with an extremely small FFT size (say, 16 or even 2), if you don't do any processing, the re-synthesis is exactly the same as the original signal! The information in frequency domain is the same as the information in the source signal, and as long as you don't change it, all of this information is back to the time domain after an inverse FFT. Now, if you do process the sound, things are different.
The Forbidden planet example is very interesting, especially if you pair it with Mark Dolson's Computer Music Journal article "The Phase Vocoder: a Tutorial". Just a bunch of bandpass filters.


And thanks for the tutorials! The FFT is indeed a magical thing.

Thanks a lot for the reply! 
And thanks for the tutorials! The FFT is indeed a magical thing.

Basic FFT question