Why not an MP3
An MP3 is a copy.
This is the original.
Most sound NFTs are a token with a link to a compressed audio file. The token is permanent. The file is just a file. A Non-Fungible Sound works the other way round.
| Typical MP3 NFT | Non-Fungible Sound | |
|---|---|---|
| What the token points to | a file someone uploaded | a result: the engine run on this punk's traits |
| Audio quality | lossy: MP3 throws away detail to save space | rendered at full 44.1 kHz and delivered as lossless WAV |
| Can it be swapped? | yes, if whoever controls the link changes the file | no: the engine's fingerprint is fixed in the contract forever |
| Can you check it is genuine? | only if the file's hash was recorded | yes: rerun the engine, and check its hash against the contract |
| Why this sound? | because it was uploaded | because of the punk: its traits decide every part of it |
| Storage for 10,000 | about 420 MB of audio files | one 38 KB program |
| How you get one | buy it | mint it: the contract draws a random punk, you get its voice |
1 · quality
No compression, ever
An MP3 keeps the parts of a sound your ear is least likely to miss and throws the rest away. That's a sensible trade for a playlist. It's a strange one for something you're meant to own.
A Non-Fungible Sound is never compressed. The engine renders it sample by sample at 44.1 kHz, and the NFT's audio is served as lossless WAV. The source isn't a file at all: it's the program.
2 · identity
The sound is the punk
Attach an MP3 to a token and the link between them is arbitrary: any file could have gone there. Here the voice comes from the punk. The pipe is the exhale, the frown is the rasp, the rare trait is the signature. Swap in a different punk and you get a different voice. Pick the same one and you always get this voice.
3 · proof
Checkable by anyone
The contract stores the SHA-256 fingerprint of the engine, and nobody can change it, not even us. Take the engine, compute its hash, compare it with the contract, then run it on any punk. If the hash matches, you're running the recipe the collection is built on.
4 · size
38 KB for all 10,000
2.6 seconds of 128 kbps MP3 is about 42 KB. Ten thousand of them come to roughly 420 MB, all of which somebody has to host forever. The engine that makes every one of these voices is a single 38 KB file, smaller than one of those MP3s.
The honest part
What still depends on this website
The ownership record, the engine's fingerprint and the minting rules live on Robinhood Chain. The rendered picture and WAV file your wallet shows are served by this website today, from the same engine. If the site went away, the voices wouldn't: the engine can be run anywhere to render them again. The contract owner can move the metadata link to a new home (for example IPFS), but can't change the engine fingerprint.
One more detail: the engine runs in ordinary floating-point maths. Two different computers make the same voice, but the very last bit of a sample can differ from one to the other. The recipe is exact; the audio file is a faithful rendering of it. We aren't committing an audio hash on chain until every sample is bit-identical everywhere.