No, I was using [log~] for the frequency decay and [curve~] for the amplitude envelope. Those objects aren't in the heavy compiler. I just simplified things.
And yes, the OWL crashed a few times
Oh, and I noticed a strange behavior in [+] object (I suppose in all operators). With heavy compiler, this objects don't save the state of the inlets. Triggering the object won't make the last operation and you need a [i] or [f] object to save the last value.
On this patch, it's done with the [+] object before [mtof].