aceslowman

First I'll ask a few questions that relate to gen in general, then I'll explain some of the vosim specific issues I'm having:

1.With gen, what are some good methods of precisely syncing two gen patches? Both patches are identical, but both need to be triggered at the same time. What I am trying to do currently is feed in all of my (sig) parameters into each, and within, hold them back using a latch. After all have been loaded in, a click~ is triggered, which I hoped would release the parameters into the rest of the patch, and allow for a precise trigger. I'm still ending up with some inconsistencies in timing that I can't find the source of.

2.My patch needs 12 parameters, as signals, fed in at the same time, and I am currently storing them in a coll. When I go from one row in the coll to the next, there is obviously a delay of some amount. What might be a good way to reduce that delay as much as possible? Vosim has me dealing with values in microseconds, so precision is definitely my concern.

Now the vosim stuff. Currently I'm working on adapting the specs for a vosim (vocal) synth into gen~, both as a way to get more chops, and also delve a little deeper into the math related to synthesis. First of all, have any of you had some personal experience or successes with vosim as a method of vocal synthesis? I found the vosim example included with gen, but it isn't the full system. I've adapted it to include all 12 variables and have been using the test data included in some of the original authors texts, but they aren't perfect.

They are working in a broad sense, and when I am really slowing down the oscillators, things seem to be working just fine: I am terminating it after a certain number of pulses, I have another termination after a certain number of periods, etc. The main uncertainties I have relate to the random modulation of M (the pause after the number of pulses have elapsed), and mixing and syncing two vosim oscillators.

When the term "random modulation" is used, I am interpreting it two ways. One being like rand~, and the other being [noise] multiplied by a sine. The test data provided gives a modulation frequency even when the type of modulation is random, which I feel keeps me constrained to those two options.

If anyone has some advice, experience, or anything, it would all be massively helpful. Thank you. Below is my vosim patch as it is right now, included with the test data.

Now the vosim stuff. Currently I'm working on adapting the specs for a vosim (vocal) synth into gen~, both as a way to get more chops, and also delve a little deeper into the math related to synthesis. First of all, have any of you had some personal experience or successes with vosim as a method of vocal synthesis? I found the vosim example included with gen, but it isn't the full system. I've adapted it to include all 12 variables and have been using the test data included in some of the original authors texts, but they aren't perfect.
They are working in a broad sense, and when I am really slowing down the oscillators, things seem to be working just fine: I am terminating it after a certain number of pulses, I have another termination after a certain number of periods, etc. The main uncertainties I have relate to the random modulation of M (the pause after the number of pulses have elapsed), and mixing and syncing two vosim oscillators.
When the term "random modulation" is used, I am interpreting it two ways. One being like rand~, and the other being [noise] multiplied by a sine. The test data provided gives a modulation frequency even when the type of modulation is random, which I feel keeps me constrained to those two options.

If anyone has some advice, experience, or anything, it would all be massively helpful. Thank you. Below is my vosim patch as it is right now, included with the test data.


vosim-synthesis-in-gen-and-some-general-gen-questions

1. Yes detecting a signal pulse to threshold changes via latch or sah is the most general way, and preserves sample accuracy. So long as all latches use the same control trigger that should be in sync. If this isn't working, maybe the problem is outside gen.

2. Maybe try a dict? Or alternatively, store parameters in a buffer~, as 12 consecutive samples (or 12 channels of a single sample frame). That way you can read them directly and immediately within gen, no need to cross the message barrier at all.

3. Sorry I've never worked with vosim, maybe someone else here has?

ACESLOWMAN, i would be very interested in seeing the full patch (some textfiles are missing). any chance you could upload this?

As MK points out, the data is not included in the patch. Out of curiosity, what is the source of that data?

Generally, as I understand, Vosim has not proven a very functional method to achieve voice simulation. Many of the claims of Kaegi appear too optimistic, and/or are not verifiable. Maybe I mean to say: don't have too high hopes. Otherwise it is a fantastic synthesis method.

In order to build on this model, it would of course be good to write it in genexpr code directly. Some things are easier implemented that way. In terms of the randomisation, noise immediately sounds so noisy. What about a random walk; M that grows with a random value and folds between boundaries? This could also be implemented second order (the growth factor of M being a random walk). Maybe that would lead to something less noisy but not as predictable as a sine wave modulation.

I think Graham Wakefield's idea of using a data object to store parameters is a good approach. That way it is possible to switch between different sets of parameters. Something that is suggested in the MIDIM language. Have you taken a look at that. I found this:

http://quod.lib.umich.edu/cgi/p/pod/dod-idx/midim-system.pdf?c=icmc;idno=bbp2372.1983.019

I think Graham Wakefield's idea of using a data object to store parameters is a good approach. That way it is possible to switch between different sets of parameters. Something that is suggested in the MIDIM language. Have you taken a look at that. I found this:
http://quod.lib.umich.edu/cgi/p/pod/dod-idx/midim-system.pdf?c=icmc;idno=bbp2372.1983.019


Found one of your other vosim patches on github with textfiles included if anyones interested: 

https://github.com/AustinSlominski/vosim-maxmsp

I would like to make a very simple version of vosim, with maybe 2-3 parameters in gen (for use with poly~, not for speech synthesis for example). I was wondering if there was a way to strip the thing so only the most essential aspects are kept. And i have to add, i have very little experience in gen (no experience with gen code box), and in the past i have had a hard time patching from papers, so help is very welcome.

Found one of your other vosim patches on github with textfiles included if anyones interested: https://github.com/AustinSlominski/vosim-maxmsp

I would like to make a very simple version of vosim, with maybe 2-3 parameters in gen (for use with poly~, not for speech synthesis for example). I was wondering if there was a way to strip the thing so only the most essential aspects are kept. And i have to add, i have very little experience in gen (no experience with gen code box), and in the past i have had a hard time patching from papers, so help is very welcome.


I'd love to share whatever I've learned from Kaegi's papers, I'd love to see what you would come up with as well. I abandoned the vosim route, it was my first project really going off of papers like that, and I felt what JVKR said about it's functionality is true. I do think it's a really interesting and elegant way of doing basic formant synthesis though. I believe the text file you are talking about was a collection of test data that Stan Templaars included in one of the papers, it was the values required to pronounce his name. The noise generation butchered it, but it had a fun effect.

I'm currently building a synth based off of Dennis Klatt's parallel/cascade formant synth model, which feels much much much more functional.

I'll look into stripping your gen version for any unnecessary parts. I'll post here if i run into any problems

Ok i think i know what i wanna do. I wanna keep T as a fixed basefrequency and keep the effects of C and N (without changing the basefreq). Probably add some kind of resonance envelope for each wavecycle.

Any help on how to do this is very welcome. Especially on scaling T so that it is useful for a mtof object, and on keeping the effect of N but making sure the basefreq doesn't change

Ok i think i know what i wanna do. I wanna keep T as a fixed basefrequency and keep the effects of C and N (without changing the basefreq). Probably add some kind of resonance envelope for each wavecycle. 

Any help on how to do this is very welcome. Especially on scaling T so that it is useful for a mtof object, and on keeping the effect of N but making sure the basefreq doesn't change


vosim synthesis in gen~ (and some general gen questions)