the synth i never thought i'd make
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
https://www.kvraudio.com/product/resyn-by-xoxos
kept dsp steamrollering, out comes the phase vocoder, out comes "formant shifting" the phase vocoder .. seriously never thought i'd be handling these procedures. plus once i had that, i couldn't not add a few features.. so it does cross spectral synthesis between two wav files, stereo and hifi processing, which is extravagant .. stereo or hifi runs about 30% cpu .. single voice is 12-16% ... and cpu follows pitch in some modes.. some it doesn't.. but this is the margin of realtime performance for me.
you may want a limiter if you start doing crazy modulation (the spectral shift param is very range responsive..)
for "the best results" you really need short wavs with this build, ~1.49s or less for 11 windowing and 44.1k.... i left the option to use longer files in there, which will sound warbly and experimental.. in retrospect i ought have had an option to autolimit the wav read to the max. quality length.. i might do that next weekend..
its an entry level phase vocoder time/pitch irrelevantiser, with cepstral recombination of source and filter. drop a stereo half second of watersplash into it and stretch it out for half an hour with occasional spectral shift sweeps.
a neat toy for sound.
i should overcome my frugality and build it to handle more data, since it is only sounding well with short files. but as is, you can stretch out short phrases with very nice quality and turn any tiny bit of audio into an infinite world. if the presentation isn't entirely intuitive.
i seriously, made the one part, made the next, ate sandwich, here we are... no testing time except quick stereo monitor, filters and algs work.
kept dsp steamrollering, out comes the phase vocoder, out comes "formant shifting" the phase vocoder .. seriously never thought i'd be handling these procedures. plus once i had that, i couldn't not add a few features.. so it does cross spectral synthesis between two wav files, stereo and hifi processing, which is extravagant .. stereo or hifi runs about 30% cpu .. single voice is 12-16% ... and cpu follows pitch in some modes.. some it doesn't.. but this is the margin of realtime performance for me.
you may want a limiter if you start doing crazy modulation (the spectral shift param is very range responsive..)
for "the best results" you really need short wavs with this build, ~1.49s or less for 11 windowing and 44.1k.... i left the option to use longer files in there, which will sound warbly and experimental.. in retrospect i ought have had an option to autolimit the wav read to the max. quality length.. i might do that next weekend..
its an entry level phase vocoder time/pitch irrelevantiser, with cepstral recombination of source and filter. drop a stereo half second of watersplash into it and stretch it out for half an hour with occasional spectral shift sweeps.
a neat toy for sound.
i should overcome my frugality and build it to handle more data, since it is only sounding well with short files. but as is, you can stretch out short phrases with very nice quality and turn any tiny bit of audio into an infinite world. if the presentation isn't entirely intuitive.
i seriously, made the one part, made the next, ate sandwich, here we are... no testing time except quick stereo monitor, filters and algs work.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
- KVRian
- 541 posts since 15 Jun, 2011 from Betwixt or between
A shame the website is no more- but I for one am delighted to see you back making (and freely offering) new opportunities to explore what ones and zeroes can do with sound.
Music can no longer soothe the worried thoughts of monarchs; it can only tell you when it's time to buy margarine or copulate. -xoxos
Discontinue use if rash or irritation develops.
Discontinue use if rash or irritation develops.
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
amazed to have done this! and the things before it.
concerning recent releases - updates due for everything, in some rebuild i'd dropped the thingy to make it not calculate pitch if it doesn't change, which will save a little cpu.
kaze will get longer, smoother noise. i think its still a useful concept.
myrmidon has a "wrong mod assign list" which may mess up some attempts to assign destination.
concerning recent releases - updates due for everything, in some rebuild i'd dropped the thingy to make it not calculate pitch if it doesn't change, which will save a little cpu.
kaze will get longer, smoother noise. i think its still a useful concept.
myrmidon has a "wrong mod assign list" which may mess up some attempts to assign destination.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
-
- KVRian
- 705 posts since 17 Sep, 2007 from Planet Thanet
Ace, thanks!
- KVRian
- 690 posts since 4 Jul, 2011 from England
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
updated, zip file with a 1 in teh name now
added a 'range' button that (normally off) limits analysis to a "quality period" without needing to prep wav files.
also added to the mod source list, lfos 1 and 2 optioned with + bias, which i thought would be more interesting amt modulated for read position effects.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
i'm kicking around ideas for a heavier build that handles more file, with more interface to select portions of file, more complex cross synthesis, ability to strip source/resonance, extra mod assign. if anyone has fedback.
now that i've managed to discern these processes, i wonder why they haven't been used. all the vocal vst i heard sounded like sample packs. it would be trivial (weekend) to write a windows app that allowed you to select cepstral frames from voice recordings, and assemble them in banks in key switched groups for variable crossfade rates to produce excellent quality vocal synthesis that could be freely expanded by users. it would be sorta keen to find a 64 bit developer to do this.
now that i've managed to discern these processes, i wonder why they haven't been used. all the vocal vst i heard sounded like sample packs. it would be trivial (weekend) to write a windows app that allowed you to select cepstral frames from voice recordings, and assemble them in banks in key switched groups for variable crossfade rates to produce excellent quality vocal synthesis that could be freely expanded by users. it would be sorta keen to find a 64 bit developer to do this.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
had a look more at this,
current release has a 40mb footprint for arrays (phase, phase difference, mag of filter, mag of source, straight mag is also nice to have for convenience.. all x2 for 2nd channel or crossmod filter) which only yields ~1.18s at 44.1kHz sampling rate..
making a simple economisation i can halve that (we only use n/2 bins lazy me) but even with kicking out mag, staying below 128mb of data (imo reasonable for users, i'm new to big data synths) only extends "quality analysis time" to around 6 seconds. i'll kick teh current build around for a wile for teh mo.
current release has a 40mb footprint for arrays (phase, phase difference, mag of filter, mag of source, straight mag is also nice to have for convenience.. all x2 for 2nd channel or crossmod filter) which only yields ~1.18s at 44.1kHz sampling rate..
making a simple economisation i can halve that (we only use n/2 bins lazy me) but even with kicking out mag, staying below 128mb of data (imo reasonable for users, i'm new to big data synths) only extends "quality analysis time" to around 6 seconds. i'll kick teh current build around for a wile for teh mo.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
done 4x memory, 8x in mono, stereo with cross filtering,
changed it to run at arbitrary # of windows overlap instead of 4x and 8x, find the quality often goes up quite well at 5, which is good news since stereo 5x runs ~36% and gives about the same as 8x, whereas it used to run 30% for 4x. usually didn't see benefit in going higher, went to 17x.
at 2x sidebands are as loud as the signal, which is pretty cool. i wanted to get that in the first one.
also added better file selection.. now you get 512/1024 mono frames instead of 128, but you can dramatically drop the hop size to get quite good quality.
i wanna play around with some things for a wile (fixed a thing with nyquist rendering dropping with high pitch somewhere) eg. bin multiplying pitch shifting instead of playback speed pitch, which does sound nice. much better platform with reducing hop size to increase quality and other options.
changed it to run at arbitrary # of windows overlap instead of 4x and 8x, find the quality often goes up quite well at 5, which is good news since stereo 5x runs ~36% and gives about the same as 8x, whereas it used to run 30% for 4x. usually didn't see benefit in going higher, went to 17x.
at 2x sidebands are as loud as the signal, which is pretty cool. i wanted to get that in the first one.
also added better file selection.. now you get 512/1024 mono frames instead of 128, but you can dramatically drop the hop size to get quite good quality.
i wanna play around with some things for a wile (fixed a thing with nyquist rendering dropping with high pitch somewhere) eg. bin multiplying pitch shifting instead of playback speed pitch, which does sound nice. much better platform with reducing hop size to increase quality and other options.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
it is here
https://www.kvraudio.com/product/resyn-by-xoxos
download link is for resyn 2.0, zip includes 1.0 release as well for legacy fun.
the second audio demo demonstrates use as a sound microscope.. you can throw 0.1s of audio in this and generate 100mb of data if you like. that does make my system stagger some, waits for button clicks..
FORTUNATELY TEH THING THAT MAKS this build wonderful is the ability to tailor quality. it uses a variable overlap instead of 4x and 8x, which means you can use 2x, which aliases terribly. 5x turns out to be quite good often.
only had a quick play but full moon, sun in taurus, prolly stable I wish someone had done one of these 15 years ago eh.
https://www.kvraudio.com/product/resyn-by-xoxos
download link is for resyn 2.0, zip includes 1.0 release as well for legacy fun.
the second audio demo demonstrates use as a sound microscope.. you can throw 0.1s of audio in this and generate 100mb of data if you like. that does make my system stagger some, waits for button clicks..
FORTUNATELY TEH THING THAT MAKS this build wonderful is the ability to tailor quality. it uses a variable overlap instead of 4x and 8x, which means you can use 2x, which aliases terribly. 5x turns out to be quite good often.
only had a quick play but full moon, sun in taurus, prolly stable I wish someone had done one of these 15 years ago eh.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
- KVRAF
- 2324 posts since 16 May, 2004 from Soviet Union
Hi Xoxox
I think this question already been somewhen, but anyway.
Do you really do not want to recreate your plugins in 64bit for (c++\juce etc) ?
You made many of quite unique things in physical modelling, voice modelling.
At least selected things from your collection would be very desirable to get in popular format. Good money could be paid for it.
I think this question already been somewhen, but anyway.
Do you really do not want to recreate your plugins in 64bit for (c++\juce etc) ?
You made many of quite unique things in physical modelling, voice modelling.
At least selected things from your collection would be very desirable to get in popular format. Good money could be paid for it.
-
- Banned
- Topic Starter
- 12367 posts since 30 Apr, 2002 from i might peeramid
nah. there is a current thread in the instruments forum, "32 bit plugins you wish were 64 bit" which is about on page five with no mention of xoxos plugins. current releases are potentiated with the existant resources and activities. my present and foreseeable affairs aren't likely to avail the time, resources (eg. 64 bit OS computer), energy or attention to validate saleability. there are better ways to catalyse culture nowadaze this joint is too hot, you can't operate like that around here. the only thing good money can do is buy rifles for west papuans.
you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.
- KVRian
- 690 posts since 4 Jul, 2011 from England
-
- KVRAF
- 3481 posts since 27 Dec, 2002 from North East England
Managed to miss this. Apologies I haven't really offered feedback on the new slew of things, but I've not really felt particularly creative lately.
I'm not surprised to see the kind of memory requirements you're dealing with. I've been an active user of Composer's Desktop Project since 2006, and the disk space requirements of its phase vocoder processes are insane. I seem to remember working out that any given analysis would broadly be window size x PCM wav file size x number of channels, e.g. 8192 window on a stereo wav results in 8192 x 705kbps x 2, around 11.5 Gbps, though this was ages ago and I could be wrong. Given that CDP included these processes back in the 80s when it ran on the Atari Falcon, it just blows my mind that Trevor Wishart and co had to deal with such data rates on such limited equipment. I was convinced they must include a ton of redundancy, but I tried a few file compressors on the analyses I keep around for later use and saw only modest gains, the sort of gains you'd get by compressing a WAV.
Looking forward to giving this a shot in the coming weeks.
I'm not surprised to see the kind of memory requirements you're dealing with. I've been an active user of Composer's Desktop Project since 2006, and the disk space requirements of its phase vocoder processes are insane. I seem to remember working out that any given analysis would broadly be window size x PCM wav file size x number of channels, e.g. 8192 window on a stereo wav results in 8192 x 705kbps x 2, around 11.5 Gbps, though this was ages ago and I could be wrong. Given that CDP included these processes back in the 80s when it ran on the Atari Falcon, it just blows my mind that Trevor Wishart and co had to deal with such data rates on such limited equipment. I was convinced they must include a ton of redundancy, but I tried a few file compressors on the analyses I keep around for later use and saw only modest gains, the sort of gains you'd get by compressing a WAV.
Looking forward to giving this a shot in the coming weeks.