I suppose it's just an issue of terminology but "granular synthesis" is a huge category. Pitch shifting, grain cloud delays, time stretching, beat slicing, autotune, etc all use granular playback to synthesize an output. When it comes to audio, I have always considered it to be any method of sound generation that uses streams of audio that are triggered ( by whatever rules) as grains of whatever length that make up a new sound. With concatenation the selection and ordering of grains is based off of an analysis of the data that will be contained in those grains and storing that data (and possibly the grains themselves) in a table so you can "tell" the playback engine, play me a loud sound with no discernible pitch that has far more energy in the ultra high frequencies" and things like that.
I'm not saying that this method is the same as the many grain cloud generators out there, but the playback methodology is granular. Again not arguing that this is similar to other granular tools in any way other than selecting and playing back "grains" of audio.
Have you spoken with the developers? I don't know if I'm gonna stick with this if they aren't doing a real beta that will come with real changes leading up to V1.0.
JJ
DataMind Audio Concatenator
-
- KVRAF
- 1900 posts since 8 Jan, 2022
Yeah granular has become ambiguous in terms of its definition in today's world
Xenakis' original definition dealt with micro timing of fragments of sound (sub 100ms)
The name tends to suggest something along the lines of "small things" but you're right grains can be pretty much any length nowadays. We have grains as big as rocks now!
I suppose the point I'm trying to emphasise is that concatenation's essential component is the ordering based on some kind of shared attribute.
My problem with concatenator is that while it works and is obviously tuned to deliver a certain result it's a kind of one trick pony. You can't really go "under the hood" so to speak to adjust how the corpus or the signal driving it are configured
The red timbre, Dillon Bastan and echobit products mentioned do allow you to adjust the parameters of the concatenation itself. Red timbre's implementation is probably the most comprehensive but arguably the hardest to dial in.
Dillon Bastan's coalescence does have a few options and also allows you to recurse on training the machine learning to go a bit deeper. Another nice feature of coalescence is that it allows you to use the features within a sample so you could put an entire song into it and it will analyse that.
Echobit seems a little underdeveloped but works fine and does give you a number of different parameters to weight the signal upon. it works better than I expected, to be honest.
Personally, I'd prefer to have more access to the MIR aspects of these plugins than having a closed system like concatenator. i don't have a particular need for LFOs etc. It's the sample analysis and triggering that's interesting to me.
Xenakis' original definition dealt with micro timing of fragments of sound (sub 100ms)
The name tends to suggest something along the lines of "small things" but you're right grains can be pretty much any length nowadays. We have grains as big as rocks now!
I suppose the point I'm trying to emphasise is that concatenation's essential component is the ordering based on some kind of shared attribute.
My problem with concatenator is that while it works and is obviously tuned to deliver a certain result it's a kind of one trick pony. You can't really go "under the hood" so to speak to adjust how the corpus or the signal driving it are configured
The red timbre, Dillon Bastan and echobit products mentioned do allow you to adjust the parameters of the concatenation itself. Red timbre's implementation is probably the most comprehensive but arguably the hardest to dial in.
Dillon Bastan's coalescence does have a few options and also allows you to recurse on training the machine learning to go a bit deeper. Another nice feature of coalescence is that it allows you to use the features within a sample so you could put an entire song into it and it will analyse that.
Echobit seems a little underdeveloped but works fine and does give you a number of different parameters to weight the signal upon. it works better than I expected, to be honest.
Personally, I'd prefer to have more access to the MIR aspects of these plugins than having a closed system like concatenator. i don't have a particular need for LFOs etc. It's the sample analysis and triggering that's interesting to me.
-
- KVRAF
- 5271 posts since 2 Jul, 2005
This replaced many of my black Friday purchases. I'm good with it. Using it on my big system where I have an insane amount of samples to play with is very cool compared to my underpowered laptop that has very few sample libraries or space to add them.
I must say that it would be amazing if they would add a side chain record mode that would let you resample things from your saw directly into the data set.
I must say that it would be amazing if they would add a side chain record mode that would let you resample things from your saw directly into the data set.
Don't F**K with Mr. Zero.
-
- KVRAF
- 6078 posts since 27 Jul, 2001 from Tarpon Springs, Florida, USA
Interesting concept but too glitchy for me.
Audio Concatenator
https://datamindaudio.ai/
Graphiti
https://www.redtimbreaudio.com/product-page/graphiti
I am into ambient and orchestral music, mostly evolving and smooth music.
Are these two plugins capable of this?
Audio Concatenator
https://datamindaudio.ai/
Graphiti
https://www.redtimbreaudio.com/product-page/graphiti
I am into ambient and orchestral music, mostly evolving and smooth music.
Are these two plugins capable of this?
Last edited by Kalamata Kid on Mon Dec 09, 2024 12:16 am, edited 1 time in total.
My Studio: viewtopic.php?f=4&t=7760&p=7777146#p7777146
-
- KVRAF
- 1900 posts since 8 Jan, 2022
Depends on what you put into the corpus (sample pool) really.
I recorded 40 consecutive guitar notes, all the same pitch and used that in Graphiti and the results were great.
I think if the samples in the corpus are too diverse you'll get the that "junkyard effect".
I do think graphiti could do with some kind of filter for smoothing and reverb/delay after the output really helps.
If you use Ableton Live Dillon Bastan's coalescence M4LL device is great. A lot of control over everything and includes things like phase vocoding for spectral playback of the samples.
I recorded 40 consecutive guitar notes, all the same pitch and used that in Graphiti and the results were great.
I think if the samples in the corpus are too diverse you'll get the that "junkyard effect".
I do think graphiti could do with some kind of filter for smoothing and reverb/delay after the output really helps.
If you use Ableton Live Dillon Bastan's coalescence M4LL device is great. A lot of control over everything and includes things like phase vocoding for spectral playback of the samples.
-
- KVRAF
- 2575 posts since 25 Apr, 2009 from gone
I had an idea : what if you use your voice as an input, and then load a list of say bear samples aside ? Will your voice be replaced with bear-like tones please ?
It seems that it’s made exactly for this kind of tricks, but will it work or will you get unusable garbage ? Can you make a bear talk using this one ?
I’d love to find myself, but no demo. And the price is steep. Thanks a lot in advance to the people who use it already.
It seems that it’s made exactly for this kind of tricks, but will it work or will you get unusable garbage ? Can you make a bear talk using this one ?
I’d love to find myself, but no demo. And the price is steep. Thanks a lot in advance to the people who use it already.
-
- KVRAF
- 2066 posts since 11 Aug, 2012 from omfr morf form romf frmo
Kammerl KFingerprints has been one of my favorite novel plugin tools for a long time, and it's free if your OS/DAW still supports it: https://kammerl.de/audio/kfingerprints/
It'll do the bear-replacing-human voice thing. But it won't necessarily sound like talking unless you give it a really good corpus to work with.
It'll do the bear-replacing-human voice thing. But it won't necessarily sound like talking unless you give it a really good corpus to work with.
-
- KVRAF
- 2575 posts since 25 Apr, 2009 from gone
Thank you very much for your answer, will try for sure.yellowmix wrote: Mon Dec 09, 2024 4:49 am Kammerl KFingerprints has been one of my favorite novel plugin tools for a long time, and it's free if your OS/DAW still supports it: https://kammerl.de/audio/kfingerprints/
It'll do the bear-replacing-human voice thing. But it won't necessarily sound like talking unless you give it a really good corpus to work with.
I’m still wondering if this DataMind Concatenator would succeed or not ar the job (just so I get how it works exactly, and what to expect from it).
-
- KVRAF
- 1900 posts since 8 Jan, 2022
I decided to get this since I love anything that's new and novel. So here's a mini review.
It's definitely a mixed bag in its current state.
First of all like all concatenative samplers the end results are very much dependent on the corpus (collection of samples). So that's worth bearing in mind.
On non tonal stuff like drums/drum loop I've had pretty staggering results. I used a vocal noises based corpuses and it turned the loop into something based on those noises. It didn't sound like a regular vocal beatbox though but something much more interesting.
On things like random foley used to drive the corpus of weird stuff I got some amazing otherworldly results.
However anything pitch based was more or less useless. No matter what I tried it just doesn't seem to be able to track pitch in any meaningful way.
I loaded up various corpuses of chromatic sample banks being triggered by some super basic waveforms (saw and sine) playing a melody. The results were nowhere close to tracking the pitch.
Allegedly they will be improving the pitch tracking in future updates.
The things it does right are the speed with which it analyses the samples for the corpus. Most other concatenative samplers use neural networks for analysis but this blitzes through. I believe this aspect is the big breakthrough with this plugin.
For non tonal stuff it really does deliver. I'm definitely getting results that I couldn't get anywhere else without huge effort.
Where it falls down is tracking for pitch. I could barely get usable results even with the most ideal of settings.
It also can get pretty heavy on the CPU side relatively quickly. The things that drive CPU usage are the number of simultaneous partials (polyphony control) and the number of active particles. The problem is that both these controls are crucial to achieve accuracy/better results. I cranked both controls to the max and rendered it offline and the results were amazing but absolutely unusable in real time.
So there is that option in a push.
I've decided to hold onto it because despite the heavy price tag I can see it being very usable even in its current state and if they refine things like pitch tracking it will be a really different and powerful option in the arsenal.
It's definitely a mixed bag in its current state.
First of all like all concatenative samplers the end results are very much dependent on the corpus (collection of samples). So that's worth bearing in mind.
On non tonal stuff like drums/drum loop I've had pretty staggering results. I used a vocal noises based corpuses and it turned the loop into something based on those noises. It didn't sound like a regular vocal beatbox though but something much more interesting.
On things like random foley used to drive the corpus of weird stuff I got some amazing otherworldly results.
However anything pitch based was more or less useless. No matter what I tried it just doesn't seem to be able to track pitch in any meaningful way.
I loaded up various corpuses of chromatic sample banks being triggered by some super basic waveforms (saw and sine) playing a melody. The results were nowhere close to tracking the pitch.
Allegedly they will be improving the pitch tracking in future updates.
The things it does right are the speed with which it analyses the samples for the corpus. Most other concatenative samplers use neural networks for analysis but this blitzes through. I believe this aspect is the big breakthrough with this plugin.
For non tonal stuff it really does deliver. I'm definitely getting results that I couldn't get anywhere else without huge effort.
Where it falls down is tracking for pitch. I could barely get usable results even with the most ideal of settings.
It also can get pretty heavy on the CPU side relatively quickly. The things that drive CPU usage are the number of simultaneous partials (polyphony control) and the number of active particles. The problem is that both these controls are crucial to achieve accuracy/better results. I cranked both controls to the max and rendered it offline and the results were amazing but absolutely unusable in real time.
So there is that option in a push.
I've decided to hold onto it because despite the heavy price tag I can see it being very usable even in its current state and if they refine things like pitch tracking it will be a really different and powerful option in the arsenal.
-
Sampleconstruct Sampleconstruct https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=191286 - KVRAF
- 16760 posts since 12 Oct, 2008 from Here and there
-
Sampleconstruct Sampleconstruct https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=191286
- KVRAF
- 16760 posts since 12 Oct, 2008 from Here and there
My first encounter with Concatenator, triggering experimental cello samples with an atonal piano improv. Although the pitch detection is quite off sometimes I really like the results, pretty amazing stuff, lots of potential for further exploration.
-
- KVRist
- 476 posts since 20 Mar, 2024
Had it or a bit and love what it can do - mostly layered with other sounds rather than as a solo instrument
-
- KVRian
- 1308 posts since 30 Oct, 2003 from Pacific NW, USA
Hey Simon, are you running Concatenator on your M-processor Mac? It looks like the CPU usage is relatively high for a minimal amount of notes being played. I'm wondering if that's normal for this plug-in. Thanks in advance for any info.Sampleconstruct wrote: Sun Jan 19, 2025 3:22 pm My first encounter with Concatenator, triggering experimental cello samples with an atonal piano improv. Although the pitch detection is quite off sometimes I really like the results, pretty amazing stuff, lots of potential for further exploration.
-
Sampleconstruct Sampleconstruct https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=191286
- KVRAF
- 16760 posts since 12 Oct, 2008 from Here and there
Yes, CPU can be high depending on the resolution settings. This was on an M3 MBP.morphex wrote: Mon Jan 20, 2025 1:40 amHey Simon, are you running Concatenator on your M-processor Mac? It looks like the CPU usage is relatively high for a minimal amount of notes being played. I'm wondering if that's normal for this plug-in. Thanks in advance for any info.Sampleconstruct wrote: Sun Jan 19, 2025 3:22 pm My first encounter with Concatenator, triggering experimental cello samples with an atonal piano improv. Although the pitch detection is quite off sometimes I really like the results, pretty amazing stuff, lots of potential for further exploration.