Synthesis type for mechanical sounds

[kamalmanzukie]

I don't want to use samples because my intent is/was making a plugin. Still I always prefer purely synthetic sounds. It is just a preference that maybe originates from when I started to make music. The first software I used for music used a softsynth. After I went over to sampled based software it felt like I always was dependent on somebody else's sounds and could not tweak the sounds. Although in the end I ended up modifying the sounds. I love sound design so pure synthesis is just more fun for that, even if it's possible with sampled sounds too.

I use samples though for some sounds like drums, vocals and strings because for those sounds the sound quality is important for me. But my dream is that all sounds could be synthesized with good quality. Maybe when SSDs with super huge storage sizes become cheap, I wont care if I use samples for real instruments.

I used to go through this philosophical hypothesis briefly and it was one of the contributing factors that made me go study math in a university.

As said above, the "Designing Sound" book is state-of-the-art literature on this subject matter. However, by viewing just that book, one can also come to understand the limitations of sound synthesis. In theory one could synthesize any sound if one knew how to construct its Fourier series. Another method is so called numerical simulation in which the idea is to synthesize the wanted sound by modelling its physics. In theory also one could synthesize any sound if one knew how to combine other sounds to form the final sound.

However, in practice Fourier series is often used for resynthesis. I.e. when one starts with a sample and wants to give it a "synthetic" representation. And in practice no-one ever synthesizes any sound by combining existing audio. So there are limitations to this "any sound can be synthesized" thinking.

In real world audio one ought to understand what methods are best suited for what tasks.

what i find interesting is how similar two sounds can appear in a spectrogram, but sound very different. the same overtones are there but small differences visually that you wouldnt fail to hear. in the voice especially, isolate a single harmonic in a vowel and you can often identify it. almost like every part of the sound has an inprint from the whole. speech is probably the best proof of this, a model will get you 80 percent there, you can do it with a model if you artisan handcraft a two second clip, text to speech still is plagued by the 'i have no soul' accent

[kamalmanzukie]

disagree.

you disagree with what, exactly? that the book reflects the state of the art? that the state of procedural audio has moved on in 7 years? that there are more game middleware products using procedural audio?

most of the theory for this stuff has been around forever. what new technique showed up in the last seven years?

Most of the theory of any audio dsp has been around for exactly the same forever. Funny that we've seen significant advances in eg analog filter modelling in the same seven year period we're talking about, though.
'New techniques' is a strawman, I made no such reference. Advancing procedural synthesis isnt predicated on 'new techniques' any more than advancing analog filter modelling was.

Feel free to provide evidence that consumer-targetted procedural audio isnt getting better and isnt becoming more and more utilised, though.

i wouldn't argue that things haven't improved. but how much of it is stuff that hadn't been written about in print seven years ago? i don't know if i could safely name one thing. zero delay feedback filters are recentish but wave digital filters can do delay free loops w/ nonlinearity and they've been around since the sixties. mo f**kers were phase vocoding in the sixties. anything procedural or gaming is out of my realm of knowledge though. what do they got. i wouldnt mind being proven wrong that everything is invented

[GaryG]

Who was that guy, here, a few years back who had the engine emulation in synthedit? (synthscience or something like that?) That was pretty darn realistic.

[soundmodel]

This is also pretty good resource, it's some sort of physical modelling, but there are a lot of techniques for that:

https://github.com/JSandusky/phya-code/tree/master/phya

[fluffy_little_something]

Funny title, I thought it was about the sounds a mechanic makes

[Xenobioz]

what i find interesting is how similar two sounds can appear in a spectrogram, but sound very different. the same overtones are there but small differences visually that you wouldnt fail to hear. in the voice especially, isolate a single harmonic in a vowel and you can often identify it. almost like every part of the sound has an inprint from the whole. speech is probably the best proof of this, a model will get you 80 percent there, you can do it with a model if you artisan handcraft a two second clip, text to speech still is plagued by the 'i have no soul' accent

I've been thinking about how difficult is to show on a paper if a sound has certain characteristics. And if even spectrograms are not sufficient then it's going to be really difficult to prove some sound results on a paper.
For example "this type of synthesis can make realistic bird sounds, as can be seen in the picture". Probably very few people could verify that by looking at the spectrogram. So then you are forced to prove it with a survey instead, were people listen to the sound. In your report you should then write 99% of people thought it was a bird.

What do you think, Is this true?

[Xenobioz]

Funny title, I thought it was about the sounds a mechanic makes

I've fixed it now

[kamalmanzukie]

what i find interesting is how similar two sounds can appear in a spectrogram, but sound very different. the same overtones are there but small differences visually that you wouldnt fail to hear. in the voice especially, isolate a single harmonic in a vowel and you can often identify it. almost like every part of the sound has an inprint from the whole. speech is probably the best proof of this, a model will get you 80 percent there, you can do it with a model if you artisan handcraft a two second clip, text to speech still is plagued by the 'i have no soul' accent

I've been thinking about how difficult is to show on a paper if a sound has certain characteristics. And if even spectrograms are not sufficient then it's going to be really difficult to prove some sound results on a paper.
For example "this type of synthesis can make realistic bird sounds, as can be seen in the picture". Probably very few people could verify that by looking at the spectrogram. So then you are forced to prove it with a survey instead, were people listen to the sound. In your report you should then write 99% of people thought it was a bird.

What do you think, Is this true?

i think what you are asking is if such an approach would be a good compromise in a marketing situation to promote your blue jay song (aka ear rape) vsti? i can't think of an alternative, frequency domain as visual input is no good, so, if it must be on paper, all you really got left is words to make your case. maybe some kind of graphical representation of some aspect of the sound like those snake oil audiophile products. if i had any means to just avoid the problem and use and actual sample of the audio, that's what i would do