Hi all! I don't know if this has been covered in depth yet, but I thought I'd post some findings and thoughts on the MT-32 synthesis engine.

Aliasing:
If you generate a waveform with a lot of aliasing which goes far below the cutoff point of the lowpass filter, it can be heard when the filter is low. In the MT-32's case, however, the waveform starts with no or little aliasing, and aliasing is audible only as the filter opens up. I have sound samples of this I can upload, comparing the MT-32 against Alpha 3.

Lack of filter flexibility:
The filter can't filter PCM samples, which is what someone on another page pointed out. Also, the squarewave used seems to be a pretty literal digital squarewave from what I've seen-- very sharp corners. If they had been able to use a real filter, I would expect they would have used a table-driven oscillator at least to minimize aliasing. Does anyone know if a squarewave appears in the ROM anywhere?

These two things make me think that the MT-32 is generating a squarewave with few harmonics, rather than generating a rich squarewave and filtering out frequencies.

Things to look for:
I assume that the resonance is created by adding a decaying sinusoid in whenever needed. This can be done CZ-style, by having a sine oscillator with sync and windowing it with a ramp down. What would be a dead giveaway is if there are flaws in the resonance emulation, e.g. it's adding frequencies which weren't in the original signal, the pitch maybe isn't sync'd exactly, or maybe it does not respond to PWM accurately. I also wonder if the filter on the MT-32 is symmetric-- if it's not, that seems likely to me that it would be faked. If we can find one of these symptoms, that might help.

I have sound examples from a synth I wrote which generates a rounded squarewave to simulate a lowpass and then adds resonance which I could post...

back to playing with my mt-32 =)