GVA1994

I have a .wav file which has three similar sounds seperated by a few seconds loaded into a buffer connected to wave~ (driven by phasor~). On bang I'd like wave~ to play the file from the beginning of any of the three sounds in stead of beginning halfway through one. I know I could find these points manually but I'd like to apply this system not to just this file, but to all kinds of sounds. Is there any way I can look for these points, or set these points automatically? Otherwise I'll have to render the individual sounds I guess.

Hi all,
I have a .wav file which has three similar sounds seperated by a few seconds loaded into a buffer connected to wave~ (driven by phas

findingsetting-points-in-file-to-start-playing-from

Without seeing the patch, it's hard to guess as to how many solution might be applicable for your needs.... here's some ideas, though:

-you can make sure that every file that's organized in this way will always start and stop the sounds within the file at the exact same fractional lengths(for example, personally, if i were working with a system like this, i'd just make sure every sound was at equidistant length so that no matter the length of files and sounds, if i address buffer by fractional-decimal points(ex. 0., 0.3333, 0.6666) then files and sounds can vary in length but my playback indexing will always find the sounds appropriately. 

-if the sounds within the file are separated by silence you can find the silence vs. sound by using envelope-following and detect where something starts then start it from there(using uzi and peek~ you can set this up in a non-realtime/instant-analysis sort of way so that you won't have to delay trying to read and analyze the buffer~ in real-time).... this last solution requires a bit more work and study(lookup 'envelope following'), unfortunately, i don't have time to show you this from scratch, but if you try it out and come up with something that's starting to look close to it, feel free to post back again and i'll keep an eye out to help again, or maybe someone else can help you further from there.

Without seeing the patch, it's hard to guess as to how many solution might be applicable for your needs.... here's some ideas, though:

 -you can make sure that every file that's organized in this way will always start and stop the sounds within the file at the exact same fractional lengths(for example, personally, if i were working with a system like this, i'd just make sure every sound was at equidistant length so that no matter the length of files and sounds, if i address buffer by fractional-decimal points(ex. 0., 0.3333, 0.6666) then files and sounds can vary in length but my playback indexing will always find the sounds appropriately.  

-if the sounds within the file are separated by silence you can find the silence vs. sound by using envelope-following and detect where something starts then start it from there(using uzi and peek~ you can set this up in a non-realtime/instant-analysis sort of way so that you won't have to delay trying to read and analyze the buffer~ in real-time).... this last solution requires a bit more work and study(lookup 'envelope following'), unfortunately, i don't have time to show you this from scratch, but if you try it out and come up with something that's starting to look close to it, feel free to post back again and i'll keep an eye out to help again, or maybe someone else can help you further from there.
hope it all helps.  

Thanks for the elaborate help! The sounds are separated by silence, so I'll look into the second option. I'll get back here when I have something to share. Thanks a lot!

I've looked into envelope following and have set up a bit of code that will go through all the samples and look up its value.. However how would I go about then finding these points where it goes from silence to the sound? Loop through all samples and look for a delta? Details are appreciated I'm not very experienced.

Also not sure if it's entirely healthy to bang 100k times through Uzi?

afbeelding.png

I've looked into envelope following and have set up a bit of code that will go through all the samples and look up its value.. However how would I go about then finding these points where it goes from silence to the sound? Loop through all samples and look for a delta? Details are appreciated I'm not very experienced. 

here ya go, googling "envelope follower algorithm" got me to this awesomeness:

https://pdfs.semanticscholar.org/401b/10f76b1a9b668a5df0a829873aaf356ac27f.pdf

basically, you're trying to get an outline of the waveform(rectified.... which is what you have so far), then smooth it out a bit(this helps minimize false-detection due to zero-crossings and other weirdness... i like how rms does this as it is sharper in detection but even simple low-pass filtering with onepole~ for smoothing can give great efficient results), and finally, detect for the signal to drop to or below a certain threshold(in your case, it's probably a bit simpler if you know you have absolute silence, to simply try to detect for drops to and from 0... but if you had a small noise floor then you could use a variable threshold parameter(other-than/above 0)).

if you want to get fancy, i like the rms method they describe there(i've used it before and i set the vector size to rms 24-64 samps at a time and it made for good response)... but you probably don't need that..

for what you describe, maybe detect logical transitions to and from definite silence of 0, this method isn't the only way(if you have sounds with a little silence within them, you might need to smooth the envelope more or use the rms stuff mentioned above to get more detailed and manage more diverse needs), and you might need to initialize it better than i did so it doesn't print doubles, haha:

here ya go, googling "envelope follower algorithm" got me to this awesomeness:
https://pdfs.semanticscholar.org/401b/10f76b1a9b668a5df0a829873aaf356ac27f.pdf 

if you want to get fancy, i like the rms method they describe there(i've used it before and i set the vector size to rms 24-64 samps at a time and it made for good response)... but you probably don't need that..
for what you describe, maybe detect logical transitions to and from definite silence of 0, this method isn't the only way(if you have sounds with a little silence within them, you might need to smooth the envelope more or use the rms stuff mentioned above to get more detailed and manage more diverse needs), and you might need to initialize it better than i did so it doesn't print doubles, haha:


I'm back with some more questions. :) I currently have it set up like this (see patcher). MC Wrappin recommended using onepole~ or an RMS technique. I didnt' know how to use onepole~ to smooth the data, since my output from peek~ going through abs. is not a signal.

I found the easiest way to get fast results was to use slide to smooth the data, and tweaking the slide up/down values gave me 4 starting points in the buffer that corresponded to the 4 samples in my file. I then used these points as start and end points for play~ (which does mean I have a bit of silence at the end of every sample since it plays all the way through to the next starting point, also playing the silence in between). This is fine for now, with other files however it didn't return correct start/end points because the smoothing with slide doesn't go well with the slide up/down values I used to smooth the data earlier. I also don't quite understand how slide works so maybe I'm doing something wrong.

If anyone has any tips/recommendations to make this work regardless of what file I give it I would love to hear it.

I'm back with some more questions. :) I currently have it set up like this (see patcher). MC Wrappin recommended using onepole~ or an RMS technique. I didnt' know how to use onepole~  to smooth the data, since my output from peek~ going through abs. is not a signal.

I found the easiest way to get fast results was to use slide to smooth the data, and tweaking the slide up/down values gave me 4 starting points in the buffer that corresponded to the 4 samples in my file.  I then used these points as start and end points for play~ (which does mean I have a bit of silence at the end of every sample since it plays all the way through to the next starting point, also playing the silence in between). This is fine for now, with other files however it didn't return correct start/end points because the smoothing with slide doesn't go well with the slide up/down values I used to smooth the data earlier. I also don't quite understand how slide works so maybe I'm doing something wrong.

If anyone has any tips/recommendations to make this work regardless of what file I give it I would love to hear it. 

Playing files like that can work only if there is no silence in audio files.

Whatever approach one takes, there is no guarantee that

it will work ok, it depends so much on audio files, noise floor etc.

Can You explain why You want this at all ?

el.buffet external from Lyon Putpourri can list start - end points

of buffer parts, relatively reliable and way more efficient than peek.

Win version is 32 bit only, mac version does have 64 bit, but

https://cycling74.com/forums/buf-op-on-a-long-buffer

last uploaded mxo would do what You want.

if You want to ty it I can upload maxpatch that does it.

Playing files like that can work only if there is no silence in audio files.
Whatever approach one takes, there is no guarantee that
it will work ok, it depends so much on audio files, noise floor etc.
Can You explain why You want this at all ?
------------
el.buffet external from Lyon Putpourri can list start - end points
of buffer parts, relatively reliable and way more efficient than peek.
Are You on mac or windows ?
Need 64 bit version ?
Win version is 32 bit only, mac version does have 64 bit, but
official release did not work at all.
see here :
https://cycling74.com/forums/buf-op-on-a-long-buffer
last uploaded mxo would do what You want.
if You want to ty it I can upload maxpatch that does it.


Thanks for the reply. I'm doing a project making procedural audio for games. I'm using library sounds as my source. Often these files contain a couple of variations of a sound. It'd be really great for me if I can just import a file and have the individual samples ready automatically; this greatly speeds up my workflow.

I'm on Windows running 32-bit Max at the moment. I'll look into the link you sent me. And feel free to help me out with an example patch :) Thanks!

https://github.com/ericlyon/LyonPotpourri-MaxMSP/blob/master/LyonPotpourri2.0/Windows/LyonPotpourri2.0_Win_MaxMSP.zip

buffet outputs a list of detected start-stop events.

I've sliced them as pairs and added 3rd value - length.

That can be used to trigger playback for play~

buffet outputs a list of detected start-stop events.
I've sliced them as pairs and added 3rd value - length.
That can be used to trigger playback for play~



hey, i like where this is going so i don't mean to intrude too much(please keep going with Source Audio's suggestions, it might be a faster way to get where you want), but in my original patch, i put a 'deferlow' after the uzi and forgot to explain that that was in answer to the question: "Also not sure if it's entirely healthy to bang 100k times through Uzi?"

(basically, it's 'ok' to do this when testing stuff, but in a larger set of functions within an app, your instincts were correct and you do want to be more careful(in a video-game especially, if the sound-engine is run on the same computer, there may be moments where some other process will be intensive and need the CPU, so a deferlow after the uzi can help distribute the processing power a bit more over time(by passing it to the low priority queue and allowing anything higher-priority to take precedence if needed))

also, there is one more way to do what you want, if you know you'll have absolute silence, you can just detect a sequence of 0s for start and ends(for example, a sequence of 0s(silence) over 16 consecutive samples that comes after sound would signify the end point and a sequence of 16 0s that comes before sound would signify the start point, if you'll have small durations of silence 

 the sounds themselves, then you might need to search for a longer duration of silence falling 

 them, but hopefully you get the general idea). i just mention this, because with any envelope-following method, you will have a level of inaccuracy to compensate for because the smoothing function causes some delay in response(most of the time, it's just a small error, that you can take care of with further windowing to tame clicks/pops, but some times, if you have a quiet gradual start or finish to a sound, it might become more inaccurate). anyways, try the el.buffet~ way(Eric Lyon's objects are some of the best ever! :D).... hopefully that'll work perfectly, but if you still run into troubles and want to try this consecutive-samples-of-silence analysis thing, just post here and i'll come back again to help.

one last thing, smoothing functions don't have to run so long(you have a slide~ that's somewhere around 10000 samples long? you could set that as low as 44 samples(slide~ 44 44), at a sampling rate of 44.1kHz, this gives you a response time of around 1 millisecond(time it takes to slide/ramp to the destination), if that's too tight, adjust up from there(i think i've gone down as minimally to about 16 samples of smoothing, but i wasn't using slide, i used a general equation in my own external:

https://en.wikipedia.org/wiki/Linear_interpolation

 .... and that reminds me: there is also the rampsmooth~ object which works more linearly, worth getting to know).

Best of luck! (i feel like guessing here: are you using Unity and having it talk to Max/MSP? :D)

hey, i like where this is going so i don't mean to intrude too much(please keep going with Source Audio's suggestions, it might be a faster way to get where you want), but in my original patch, i put a 'deferlow' after the uzi and forgot to explain that that was in answer to the question: "Also not sure if it's entirely healthy to bang 100k times through Uzi?"
(basically, it's 'ok' to do this when testing stuff, but in a larger set of functions within an app, your instincts were correct and you do want to be more careful(in a video-game especially, if the sound-engine is run on the same computer, there may be moments where some other process will be intensive and need the CPU, so a deferlow after the uzi can help distribute the processing power a bit more over time(by passing it to the low priority queue and allowing anything higher-priority to take precedence if needed))

also, there is one more way to do what you want, if you know you'll have absolute silence, you can just detect a sequence of 0s for start and ends(for example, a sequence of 0s(silence) over 16 consecutive samples that comes after sound would signify the end point and a sequence of 16 0s that comes before sound would signify the start point, if you'll have small durations of silence within the sounds themselves, then you might need to search for a longer duration of silence falling between them, but hopefully you get the general idea). i just mention this, because with any envelope-following method, you will have a level of inaccuracy to compensate for because the smoothing function causes some delay in response(most of the time, it's just a small error, that you can take care of with further windowing to tame clicks/pops, but some times, if you have a quiet gradual start or finish to a sound, it might become more inaccurate). anyways, try the el.buffet~ way(Eric Lyon's objects are some of the best ever! :D).... hopefully that'll work perfectly, but if you still run into troubles and want to try this consecutive-samples-of-silence analysis thing, just post here and i'll come back again to help.

one last thing, smoothing functions don't have to run so long(you have a slide~ that's somewhere around 10000 samples long? you could set that as low as 44 samples(slide~ 44 44), at a sampling rate of 44.1kHz, this gives you a response time of around 1 millisecond(time it takes to slide/ramp to the destination), if that's too tight, adjust up from there(i think i've gone down as minimally to about 16 samples of smoothing, but i wasn't using slide, i used a general equation in my own external:
https://en.wikipedia.org/wiki/Linear_interpolation .... and that reminds me: there is also the rampsmooth~ object which works more linearly, worth getting to know).

Best of luck! (i feel like guessing here: are you using Unity and having it talk to Max/MSP? :D)

Thanks a lot all!! Sorry for the late response. Yes we're using Unity with Max! I'm going to look into these suggestions and get back here with solution/more questions. Thanks!!

Finding/setting points in file to start playing from