i know how to end aliasing

[Benedict]

UPDATE:

I see you have a site that shows some of your logic: http://soundemote.io

s = sample rate, r = oversampling ratio. Nothing in this engine is allowed to move faster than half of that. Anti-aliasing isn't a filter you bolt on afterward — it's obeying the speed limit in the first place.

Ok, I see the logic in simple terms; this is somewhat like Band Limiting, in that you tell the system that anything above x is ignored QED: nothing to bounce. From my very limited grock of DSP I think you are doing the same, only perhaps at a more granular level: every event is asking the Q rather than taking the whole and saying 'ignore over x'. Assuming CPUsage doesn't become unwieldy, it may do it. Or it may simply make an FM like "1x56" behave like "1" which, while solving the Nyquistian "grunge", didn't deliver the requested result - which, in theory, oversampling should (but sadly nothing really does).

[antto]

unlimited detail, infinite detail. i like the word infinite because unlimited implies there was a limit in the first place. infinite is more exact, it describes a number: ∞ or we can drop the whole infinite/unlimited thing and just say:

"it's a chaotic equation that i tune not to alias"

ill say that moving forward.

also, this requires you to abandon classical audio synthesis, so in other words, I'm only ending aliasing in my simulated universe rather than for arbitrary audio. you have to actually jump in to get the benefit, tradeoff is you are not allowed to calculate sin() in this universe, you have to tap into the "free energy" source of "spacetime fluctuations" (i.e. the sinewave oscillating over time and the noise floor)

the reason i say "ending aliasing for all" is because i am positioned to grab a lot of the audio market, they will use my engine for their audio, be it a game, an audio plugin, etc... hmm, ok that's too optimistic. my engine will be used alongside classical synthesis im sure.

classical synthesis: good for getting something exact
chaotic synthesis: good if you don't need exact (especially organic / foley sounds).

so many buzz words, buzzwordoverflow

so maybe change your original thread title then, otherwise this is clickbait i'd say
i don't know what "universe" you live in (i have a suspicion but i'll keep it to myself for now), but a bunch of the devs who would look at your thread title, would click on it interested to learn how to make antialiased "exact" waveforms, not some uncontrollable crap, EHM, i meant "chaos"

well, congratulations, you (once again) grabbed the attention of a lot of devs here... and now what?
i'm not good at coding and math, but i don't see anything here from you, just claims and promises entangled with buzz words which don't tell me anything "exact" about your "algorithm", but it does tell me something about you.

---
mystran: yes, that's the one. i periodically eyeballed it for a few years after i first saw it but nothing was happening and i forgot it.. so it actually did have something, huh, bad targetting and bad communication maybe? any way, i've not been following gamedev stuff so, sorry for the bad example then.

[Architeuthis]

a) show it working,

i did
http://soundemote.io <- explore and learn
http://soundemote.io/sandbox <- make something (not everything works yet)

b) keep quiet til you have the next (real) game changer, or

ive already changed the game

c) ?

you can use it or not. the only thing missing is a demonstration you can understand.

[BertKoor]

Keeping aliasing out of synthesized waveforms is a solved problem.

Now how do you tackle it for distortion / overdrive / saturation / clipping ?

[Architeuthis]

↝ distortion: noise floor eats distortion like dithering for digital audio, that's our free energy source doing work (noise floor is adjustable and dictates CPU usage / fidelity), and you can amplify this noise floor for noise things. the noise should be natural-sounding (untested) because only `under-nyquist` noise gets through. the noise is filtered / split into above and below nyquist, the above nyquist data goes into imaginary land as a secret code (i just throw it away), and some information is lost to the black hole because of nonlinear filtering.

new idea: use a bandpass filter for the noise and make it adjustable, that allows fine tuning for DC vs AC and hi fi vs lo fi (untested)

wait: is it the above nyquist noise that prevents distortion and what you hear is the below nyquist noise? but then you would be hearing above nyquist noise... but high frequency noise is above human hearing... oh no... i don't know the answer (needs testing)... but then again, the entire noise signal is inside the circuit... but it's being filtered... so... the distortion also gets filtered?

im almost at the point where i can test all this... alright ill make sure i have something today.

↝ overdrive: it overdrives beautifully (saturation is just a part of the chaotic signal, there is no literal saturation unless you throw in a soft clipper... that might introduce aliasing... this is where fine tuning comes in)

↝ saturation: is saturates beautifully (overdrive and saturation are equivalent in this context: something becoming less round)

↝ clipping: clipping ends the universe simulation... well i need to implement this feature: if a signal disobeys -1 to +1 then a wire will break where it occurred... except... wait... there may not be wires sometimes... so in that case, yes, it ends the universe simulation similar to a circuit breaker function. this feature will be optional, but ill encourage keeping it on.

[Architeuthis]

got this from gemini, an equation for calculating an nth digit predictably, but digits are also very random. the point is: i need an algorithm that we can check where we are in time based on digit and comparing surrounding digits, and at the same time, use that as a noise floor. so we keep time and noise in the same instance. i could then derive a sinewave out of that noise with a bandpass filter.

before my idea was: generate sine wave, use noise to cover distortion and keep time. but this might be better? i think there are multiple "universe algorithms" that we can go with, perhaps each one will have a different overall sound quality.

Code: Select all

import hashlib

def get_irrational_noise(n: int) -> float:
    """
    Generates a deterministic pseudo-random noise value between -1.0 and +1.0
    for any integer n, representing digits of a chaotic irrational number.
    """
    # 1. Convert the position index to a string and hash it
    encoded_input = str(n).encode('utf-8')
    hash_hex = hashlib.sha256(encoded_input).hexdigest()
    
    # 2. Extract a slice of the hash (first 8 hex chars) and convert to integer
    # 8 hex characters give us 16^8 (4,294,967,296) possible discrete values
    hex_chunk = hash_hex[:8]
    integer_value = int(hex_chunk, 16)
    max_possible_value = 16**8 - 1
    
    # 3. Normalize to a 0.0 to 1.0 range
    zero_to_one = integer_value / max_possible_value
    
    # 4. Map linearly from [0, 1] to [-1, 1]
    noise_value = (zero_to_one * 2.0) - 1.0
    
    return noise_value

# --- Example Usage ---
if __name__ == "__main__":
    print("Sequential samples (simulating a timeline or stream):")
    for i in range(5):
        print(f"Index {i:2d} -> Noise: {get_irrational_noise(i):.6f}")
        
    print("\nInstant infinite lookup (jumping straight to distant indices):")
    print(f"Index 999999 -> Noise: {get_irrational_noise(999999):.6f}")
    print(f"Index 1000000000 -> Noise: {get_irrational_noise(1000000000):.6f}")

wait, is the rule that: pure random turns into pure order with a filter? pure order is undistorted by pure random because you can't calculate pure order as that requires infinite energy, but you can calculate pure random..? or at least it's easier than pure order??? or is it the same difficulty?

1. do we generate noise and filter sine? or
2. do we generate sine with dithering and extract meaning? or
3. do we generate sine and noise at the same time? or
4. do we use a chaos generator and skip all the above? is that even possible?

all these questions need testing. im starting with #2.

#1disqualified due to the nature of filtering introducing delay, but it may still be useful to experiment, it may offer a totally different universe of sound. wait, but you could compensate for that delay with prediction or just have few samples of latency, this could turn out to be the best solution paradoxically, but at the cost of latency... or im just reinventing linear filtering. psuedorealtime linear filtering

[Architeuthis]

serious question: is it possible, with some latency, to remove discontinuities by filtering backwards and forwards in time? the challenge: the fact you have to go backwards in time. the upside: we only have to do it for a few samples (i think, or that absolutely does not work whatsoever), but perhaps it could be reserved for offline mode if all else fails (for a last bit of linear filtering clarity/perfection).

wait... this may be touching upon resynthesis and signal reconstruction.

damn, this might be the start of the vector audio resynthesis engine.

[Architeuthis]

i said i would give you proofs of concepts today, but i need a break from this. so here's just a random video that hints at possibilities. notice that you can see discontinuities as invisible lines, it will help me debug them in the future.


(for some reason out back into freq is not working, need to debug that)

[antto]

blackholes, noisefloor, dither, free energy, universes, "wires breaking", "get_irrational_noise"

you can use it or not. the only thing missing is a demonstration you can understand.

yeah, i'm kinda dumb
what i understood so far is that maybe you're chatting with a chatbot for long enough that you two slowly developed your own pseudo-scientific language where when you say "send the aliasing into a blackhole" the chatbot "thinks" (hallucinates) one thing, while you think a completely different thing, but it doesn't matter as long as the conversation keeps going

this is as far as i can understand with my limited view

i clicked for 2 minutes on your website:
i think i found some pixels which forgot to go into the blackhole