About Emulation

As we said:

the TransFormSynth is not a sample-playback synthesizer.

But if you have your heart set on it, it is often possible to make it play back a sound so that the resynthesis approximately reproduces the original sound. There are several things to do: some are straightforward and some take a bit more finesse. Let's start with the easy ones:

First, suppose that the original sound has harmonic overtones. Set the 2-d joystick position to maximize tonalness (i.e., make it as harmonic as possible and do not attempt to temper the partials). Or, if you are using the linear keyboard, set the "harmonicity" slider to harmonic (i.e., no tempering).

Second, set the depth parameters to 0 since the original sound has no formant filtering and no harmonic modulation (and no attack modulation). Make sure that the attack randomize is set to 0. The original sound has no modulations or randomizations on the noise either.

Set the time base dial to 1. The original sound proceeds at the rate of one second every second.

Set the attack/noise dial to 1. The original sound has noise equal to one times its noise component.

The values for the envelope should be set (in the 'patch data' table) to: 2 ms attack (any less and you may get a click), 500 ms decay, 1. sustain factor, and 30 ms release. These guarantee that the envelope has (almost) no effect.

Make sure the slew time is short -- say 2 ms.

Now for the trickier ones: the number of peaks parameter (dial 1) and the peak detection threshold (dial 3). The goal here is to choose to resynthesize using the number of peaks that the sound actually has. So -- play with the detection threshold and find a value that is robust -- not too small and not too large so that the number of peaks detected (shown in the little number box 'peaks found') makes sense. A flute might have 5 to 10 peaks. An oboe might have 30 or 50. A guitar might have 50 or 60 in the low register and 20 or 30 in the high. Use your judgement. Then set dial 1 to the same number as you have found so that it resynthesizes the same number as it detects.

So do you see why it's tricky? If you set the threshold wrong (say too small) then you detect more peaks than are actually there. In the resynthesis, you then treat all of these incorrect "peaks" as if they were real harmonics of the sound (in reality, they are part of the noise and should be treated as such). On the other hand, if you set the detection threshold too high, then you don't detect all the real peaks and when you resynthesize, you are treating true peaks as if they were noise (in reality, they are harmonics and should be treated as such).

With any given sound, if you play around a bit, you can usually get the resynthesis to be approximately the same as the original, if you play it at the original pitch (the "h" key on the computer or the MIDI note corresponding to an untransposed sound). But as we said earlier: why would you go to all this trouble just to resynthesize something that you already have? I guess it's nice to know that it's possible.