Sample Vs Frequency Packets.

Started by vicktech, October 12, 2010, 21:39:31

Previous topic - Next topic

vicktech

Bad thing about samples is mangling sounding and duration when frequency changed by affecting playback speed.
Bad thing about instruments is they need many samples to supress sample mangling. And they need volume corrections. Because higher frequencies put more energy out than low ones at the same amplitude.
How about doing some RE-SEARCH in sound architecture and try to represent sample as frequency packet.
Sample will become sound processor ready. And despite wave structure will differ from original, even each time, - the sounding will be almost exact.
Frequency-energy compensation can be build-in feature of sound processor routine.

Besides sample will no more be affected by frequency change.
Instrument will obtain completely renewed meaning and become powerful and advanced as it ought be from the very beginning.

(Please no copyright claims, this is for free use)

psishock

We already have a solution for this, they are called "soft synths", no need to reinvent everything with new a new name. :D

Buzz tracker example, has an internal and simplified plugin system, that includes sound generators. But we have a very nice VSTi support in OMPT, and you can find many powerful and free ones, that you can use for your projects.

Take your time in this topic:
http://forum.openmpt.org/index.php?topic=332.0
I'm as calm as a synth without a player.  (Sam_Zen)

vicktech

Quote from: "psishock"We already have a solution for this, they are called "soft synths", no need to reinvent everything with new a new name. :D

Buzz tracker example, has an internal and simplified plugin system, that includes sound generators. But we have a very nice VSTi support in OMPT, and you can find many powerful and free ones, that you can use for your projects.

Take your time in this topic:
http://forum.openmpt.org/index.php?topic=332.0
This is interesting...
But I doubt it can record and playback arbitrary instrument from real world, within 15 minutes.

I will summarize my mental efforts from renoise forum:

Well..., instead of direct sampling at fixed sample rate, - frequency spectrum analyzing, and finally leasing ought be performed. Because human's ear hears sound by using over 40 000 sensors, each responding to its own frequency, (at least I know it from somewhere, if I'm not mistaken).

Then sample will consist of frequency packets and amplitudes, both represented by precise digital arithmetics.

Thus frequency can be changed with no harm to play-time of sample.

And it is followed by another idea:
For natural sounding of sample it should be represent as compresssed hologram.
The meaning is: All notes (or even smaller frequency shifts) are played and analyzed, then compressed by algorithm which rids of unnecessary redundancy. According to required frequency hologram will give corresponding frequency packets.
And specific sounds of sample, like noise from touching or scratching strings or beating at them will be transposed accordingly to hologram. So it will be reproduced correctly - no matter what note you play.

Idea is here. It can not be stopped :)

psishock

QuoteIdea is here. It can not be stopped :)
Lol, ok.

I may repeat myself here, but as i've said, this idea is already done (in different approach tho'). :D
Your answer is resynthesis, and its implemented with the VSTi technology. Trust me, you're talking with a synth/sound engineer, so i am aware of a bunch of stuffs from the field, even if im not using them actively on my projects.

I could write a number of sentences, about how it actually works, but i rather let the expert do this, with a nice, prerecorded presentation:
Check the Morphine tutorial videos from ImageLine. (particularly the part5 speaks about our term, but your should listen to all of them from the beginning, to understand all the concepts, working methods and sampling logic). Its a very powerful and serious tool already, and perhaps it will move closer and closer to "perfect" with the new versions (more harmonics, more features).

http://www.youtube.com/watch?v=jX7hIqmRJNI (part1)
http://www.youtube.com/watch?v=NA-wDS0pbyM (part2)
http://www.youtube.com/watch?v=2LisB8duRRQ (part3)
http://www.youtube.com/watch?v=UwgBNMxXQa8 (part4)
http://www.youtube.com/watch?v=KcmGCGVdnhY (part5)
http://www.youtube.com/watch?v=MzjhvXfZ_Ko (part6)
http://www.youtube.com/watch?v=5vMQTSvgyr4 (part7)
http://www.youtube.com/watch?v=JIHcKr2ilu0 (part8)
I'm as calm as a synth without a player.  (Sam_Zen)

vicktech

On my way.
1. For precise sound at least  48000 paralell generators are needed. Not 1028.
2. Noise is same sound as waves, there is no need in separate noise generator.
3. Transposition by halftone is not professional. And fine tune of 100 divisions is really lame. It should be 144 halftones in doubling frequency (octave - actually is "dozave").
4. Attack, decay sustain controlled by turning things - what is it? kinda sick joke? Imstrument can have multiple attacks, sustains and attacks again.
5. I don't believe in delay and reverberation. They only may sound dirty. True music should be done only my high skill in manipulating temp, volume, frequency, and silence.
6. It could not reproduce drums...

... more to come.

psishock

Quote1. For precise sound at least 48000 paralell generators are needed.
in theory yes, but in practice much less is enough to "fool" the ear. You wont tell the difference, encoder algorithms works on the similar practice. Also im sure you could imagine, how much processing power would 48000, realtime controlled parallel generators consume. That would be insane even for the cutting edge processors.

Quote2. Noise is same sound as waves, there is no need in separate noise generator.
Sure, but again in practice, using a special "noise" oscillator can enormously free up the processing power, because you are using 1 generator, instead of many hundreds, to archive that "complex" sounding.

Quote3. Transposition by halftone is not professional. And fine tune of 100 divisions is really lame. It should be 144 halftones in doubling frequency (octave - actually is "dozave").
fair enough, but that initial halftone/finetune setup is just for importing the sample around c5. You can play with every harmonics and their precise pitch afterward in the editor. Surely, they could add even more precision (that may happen over time), but i think, that this level of accuracy is sufficient already for most projects.

Quote4. Attack, decay sustain controlled by turning things - what is it? kinda sick joke? Imstrument can have multiple attacks, sustains and attacks again.
relax, that is just a basic, global ADSR control on the top of everything. :D You can also use precise envelopes, that can be routed to your desired control, to modify your soundings. Check the modulation section.

Quote5. I don't believe in delay and reverberation. They only may sound dirty. True music should be done only my high skill in manipulating temp, volume, frequency, and silence.
that is your own choice, naturally. You can reproduce of course everything hand-by-hand, but effects are there to take a big amount of producing workarounds and tricks from your back. You know how you want it to sound, and with the proper effects, you can archive it without long, bothersome workarounds/manipulating. Even so, some specialized effects almost cannot be reproduced by hand, because of their complex algorithms and host program's sound manipulation limitations.

Quote6. It could not reproduce drums...
why not? Personally i havent tried it myself, but what i've read from it, any type of loaded samples can be reproduced with the resynthesis, the technology is not limited to "some type" of sounds.
I'm as calm as a synth without a player.  (Sam_Zen)

vicktech

Quote from: "psishock"
Quote1. For precise sound at least 48000 paralell generators are needed.
in theory yes, but in practice much less is enough to "fool" the ear. You wont tell the difference, encoder algorithms works on the similar practice. Also im sure you could imagine, how much processing power would 48000, realtime controlled parallel generators consume. That would be insane even for the cutting edge processors.

Actually one have to do 48000*48000*2 calculations per second. Calculation may involve complex arrays and should be implemented in assembler.

Good news is that lower frequencies may be calculated with lower rates, so it will decrease required resources at least twice.
Another news: not all generators will be used constantly, - they may be allocated at request.
One more: using specialized hardware chip, job can be done with minimal resources of CPU and closer to realtime

QuoteSure, but again in practice, using a special "noise" oscillator can enormously free up the processing power, because you are using 1 generator, instead of many hundreds, to archive that "complex" sounding.

It only tells us that something is not completely right with oscillation theory.

Quote3. Transposition by halftone is not professional. And fine tune of 100 divisions is really lame. It should be 144 halftones in doubling frequency (octave - actually is "dozave").
fair enough, but that initial halftone/finetune setup is just for importing the sample around c5. You can play with every harmonics and their precise pitch afterward in the editor. Surely, they could add even more precision (that may happen over time), but i think, that this level of accuracy is sufficient already for most projects.[/quote]

Not, you don't understand.
Such numbers like 100 are not harmonic and will result in "beating" of frequencies when sound is mixed with other sounds.
So it is quite a mistake of developers.

psishock

QuoteActually one have to do 48000*48000*2 calculations per second. Calculation may involve complex arrays and should be implemented in assembler.

Good news is that lower frequencies may be calculated with lower rates, so it will decrease required resources at least twice.
Another news: not all generators will be used constantly, - they may be allocated at request.
One more: using specialized hardware chip, job can be done with minimal resources of CPU and closer to realtime
designing and projecting a new, specialized hardware architecture is really out of my league. In theory, 3d graphic could be already perfect with a well designed chip, that allows unlimited details on pixel precision field, that would be only calculated and rendered realtime, on the user's visible area. We havent seen even a trace of that technology yet however. Maybe there are people projecting both of them already as we speak. I could easily imagine that many of the musicians/sound engineers brainstormed about this "sound reproducing hardware" solution. I think we should only talk about present solutions, who knows where will we all be in about 10-20 years. :D

QuoteIt only tells us that something is not completely right with oscillation theory.
well, it works on simpler (again, this is quite relative) sounds, and it works on universal processors. We can compose a whole piece of musics and still able to edit/listen to them in real time. I am here to make cool songs and to help people with working solutions today, also im fairly satisfied with the present tools. Surely, in future everything will be more and more simpler to use. Nowdays vstis/hardwares are a lot more powerful already, than the ones from around 2000.

QuoteNot, you don't understand.
Such numbers like 100 are not harmonic and will result in "beating" of frequencies when sound is mixed with other sounds.
So it is a quite mistake of developers.
Oh, now i see what did you mean. This can be troublesome indeed on some cases. Well they may had a pretty good reason for this, or it has been some kind of a technical limitation. Im sure that the experienced designers, that cared for all the important details, would not miss this one by any kind of accident.
I'm as calm as a synth without a player.  (Sam_Zen)

vicktech

Finally I see that Morphine is good try to make us happy,
Bat what I offer is much more proper and precise.

Rxn

vicktech, I am not sure if you are not able to explain yourself, but generally
it seems to me that you have no slightest concept of what digital audio is.

I've been into the area of digital audio since 1997 and I am a qualified
sound engineer that is not to show off but rather explain reasons behind my
opinion expression.

You've got to polish up a bit on the topic, there is plenty of information
available on the Internet these days, not at all like back in 1997. Afterwards
you should give it a try again, you might be onto something.

vicktech

Quote from: "Rxn"
You've got to polish up a bit on the topic, there is plenty of information
available on the Internet these days, not at all like back in 1997. Afterwards
you should give it a try again, you might be onto something.

It's a big luck to meet sound engineer!
I will not deny my small amount of knowledge in digital sound processing.
As user i can tell only one thing: I am not satisfied by instruments used in trackers. Instrument should be realtime, small, it must sound perfect - we use computers, after all.

I might be onto something? I do not understand what do you mean.
Everything i saw is eating tons of resources and still no satisfaction.
There are other engineers at renoise forum, one of them made concept of matrix transformation, basing on my idea. You can have a look.

Rxn

What you are talking about here sounds like what I call "spectrum synthesis"
-- quite a novel area of research in digital audio.

Too bad you don't quite understand what you are talking about and how it
relates to digital audio:)

For example, when you are talking about 48000 parallels generator you are
clearly being misled by a wrong understanding of what sample rate of a
digital file is. It appears that you think that sample rate = number of basis
cosine waves a signal contains which is not true since it contains frequencies
of up to 24000 with the range of 20kHz-24kHz being an aliasing wasteland.

From this alone I'd recommend you to do some study of the current
digital audio, what you are talking about could be the next step of its
developmentbut you have a lot of work to do before getting there.

Saga Musix

Quote from: "Rxn"with the range of 20kHz-24kHz being an aliasing wasteland..
I have to keep that sound bite in mind :lol:
» No support, bug reports, feature requests via private messages - they will not be answered. Use the forums and the issue tracker so that everyone can benefit from your post.

vicktech

#13
Quote from: Rxn on October 19, 2010, 18:39:49
What you are talking about here sounds like what I call "spectrum synthesis"
-- quite a novel area of research in digital audio.

Too bad you don't quite understand what you are talking about and how it
relates to digital audio:)


Too bad - nobody really does.
Otherwise they would have succeed.

From now on I'm doing mental scan on "Quantum pulses" + Fractalization sound model.
I came to this thought from noise on TV screen.
Bad or weak signal makes system loose quantums, you know....

Noise - chaos or extremely dense information?
Got answer? You probably need to do some research of known book before you can tell. (Hopefully)
Best wishes.


Saga Musix

Quote from: vicktech on October 30, 2010, 22:42:29
Too bad - nobody really does.
Otherwise they would have succeed.
If the laws of physics were as simple as you imagine, we would have perfect pitch shifters and time stretchers today. Yet we don't but not because we haven't understood but because we can't.
» No support, bug reports, feature requests via private messages - they will not be answered. Use the forums and the issue tracker so that everyone can benefit from your post.