@lacuna Oh I see, my bad! I hear the smearing now. I wonder if increasing the overlap would help? I wonder if that would require a different shaped window? I would think it would also affect the normalization constant.

Edit: Hmm, maybe not. I increased the overlap to 32 and the normalization const to 12N but kept the Hann window and it still smears. I03.resynthesis 4096.pd But this is just me taking a random shot in the dark.

@Blair I don't really understand your question, but there's nothing about large FFT block sizes that would cause smearing, so I would check if it's something else. You could confirm my claim by converting I03.resynthesis.pd in Pd help to 4096 and compare with the original. As for making the window smaller than the block size, that would be similar to putting a tremolo on your signal before and after the frequency domain processing because the windowing is done in time domain, which you can see in that same help example. I'm guessing that's not what you want.

@lacuna said:

Katja's meter measures 1 sample latency additionally, so real latency is -1 sample.

Does that imply you just downloaded Katja's meter? If so, how? All hurleur links are broken for me.

@whale-av Thanks, I wasn't aware of any of the tools in doc. Looking forward to studying their version.

@lacuna this one https://forum.pdpatchrepo.info/topic/7710/latency-on-output-and-input/3

Is this a good round trip latency test, and should I believe the results I'm getting? Windows 10/64
RTL test.pd

Here is my MOTU ultralight mk4 with its Main L output connected to Mic/Line 1 routed to Computer 1. It's interesting that Pd is set to the lowest delay possible for a 64 sample blocksize, but the measured latency is over 5X that. About 10 years ago, @katjav posted she was getting 1.5X on Linux.

And here is my Yamaha 01v96i with Monitor Out L connected to Channel 11 line in routed to bus 1/USB 1.

Finally, is it true that I could expect these latencies if I were to use Pd as a live signal processor provided my patch doesn't introduce latency normally (e.g. uses overlapped windowing for FFT resynthesis) and that Pd is not falling behind in audio processing?

@FFW said:

As I understand ChatGPT is based on a generation which choose one word after another, it doesn't have the answer to our question until it writes it down.

To be honest, that's exactly how I wrote papers in college.

Like a lot of software problems, it seems like a lot of the difficulty is in defining the requirements. I saw that frequently when I was working.

@Kane99 Are you only interested in folks who do synthesis in Pd and who use synthesizers?

@ddw_music said:

The integral of a sine wave is always 0. The sum of the integrals of multiple sinusoidal components, then, must also be zero.

Oh duh. I guess I was hell bent on making that joke

@whale-av

All equipment manufacturers state " Maximmum RMS, or maximum peak music power (4x RMS), and no warble tones"

I can't tell if that was a joke...I've never seen that statement.

@ddw_music Definitely interesting. It's too late for me to think clearly about this, but if we stipulate that people can't hear the relative phases of harmonics, then shouldn't there be some arrangement of a given tone's harmonic's phases such that the DC component is maximal (I'm thinking of all the times I got the ramp wave wrong using additive synthesis)? And that that tone would be indistinguishable from the original? And so any sound could potentially have dangerous gear-eating DC components lurking within it? And so the best course of action if you own precious audiophile components is to never listen to anything?