pitchshifter

[pansoul]

Hello
i try to make a pitchshifter with the phasevocoder.
it works but there is a bug probably due to init conditions.
if i plug the module and especially the audio input when the usine engine is on OFF and that i put ON after, it works.
But i put ON before and the i plug i have a bug.
I put the code and the module in the uploader if you to reproduce the problem.
thanks

[martignasse]

Hi pansoul,

I didn't try to recompile, just used the bin you provided.

I cant reproduce exactly what you describe.

For me, release bin seems to work ok, but debug bin produce an extremely saturated audio :

- around 220 dB when the module is dropped in the patch ! :O
- if pitch is changed with a slider without any sound input, it can grow to 600 dB.... wow.
- when an audio input is connected, the output stay saturated until the pitch is set to positive value.

I saw nothing obvious in the source file to explain that, but it seems related to the phasevocoder in debug mode



Hope it help.

[pansoul]

yes you are right
i finished the module a few days ago and my description of the bug wasn't completly correct.

in release : no problem

debug : i have the same comportment as you describe

Does it mean that release mode must be prefered rather than debug (for such a module it takes the same time to generate)

[martignasse]

Does it mean that release mode must be prefered rather than debug (for such a module it takes the same time to generate)

Well, it essentially mean that your module has some potential bug in release mode
Even if it's not noticeable in release mode, it's a good thing to find why the debug mode produce this 'bad' result.

in short :
debug mode is for the development phase, where you can trace actions inside the module at runtime (see this wiki page).
in addition, the debug mode make a lot more verifications, about buffer overflow, memory protection...

release mode is for the final product, it's compiled without all the debug infos and with some optimisations (for speed, or size...)

about your code,
I noticed that you manually transpose the spectral FFT ( forward -> manual transpose -> inv ).
But there is a spectraltransposer::tickTransp(float semitone) function, it's maybe better to use it ?