Hi, I'm building an audio sound engine for a mobile game using libpd and the rjlib library. I want it to be CPU efficient so as to run on lower end android phones too without glitches

My first assumption is that playing back audio is less costly than using synths : is this a fact or am I wrong from the start ?

The idea is to build a flexible sampler with distinct attack, sustain and release parts. I can then load any synth sound I exported from my DAW and play it almost as if it were the original synth (without all the modulations of course). Kind of like the ableton sampler: I could choose the start, the sustain loop portion, and the release part of the sound. That would allow me to add portamento, amplitude, pitch and filter enveloppes, hold notes forever, instantly change the pitch, with no glitches between parts ... It would have to sound like I was playing the original synth preset from my DAW, inside PD !

However this looks a bit trickier than I thought and seems to require a bit of patching. In the end I'm wondering if it's not going to end up costing more resource from the CPU than using a PD synth ?

So before I start building this thing I'm wondering : is it a good idea, or I am missing something here ? And also, does it not already exist ?

I hope I've managed to be somewhat clear, i'd love to hear your feedback !

Cheers
Simon

@simonhillfr Are you developing with Unity? upvoted as Im interested in using unity3d for a pd gui designs. Not really sure how I could help but I'm really curious any game related

It depends on the method and complexity of synthesis - if you use filters, feedback and anti-aliasing filters ect.
The sampler might basicly be arrays as look-up tables < read interpolated with tabread4~ with a phasor~ < modulated with ramps and some AM.
With similar ingredients you can already patch a very basic synth but more complexity will require some more cpu.
If you want time-stretching and independent pitch-modulation, the sampler will become more complex with FFT.
(You see, in reality either one becomes a hybrid quickly: for example [cos~] is a table-lookup, too.)
Reading a complex pre-made sample will use the same CPU as a very basic sound.
A sampler will require more RAM, as the arrays are stored in memory. You can easily calculate the reqired space for this if you know samplelength, bits and samplerate.
Given the synth- or sampler-apps available for phones, I guess you can patch quite complex instruments in either way. Of course this depends on other expensive tasks running along with it.
The CPU-load depends on blocksize and latency, too.
Often in PD-Vanilla the bottleneck are the GUI elements, wich you might not use anyway.