How it works
How a punk gets its voice
Nobody sat in a studio. Each voice is computed from the punk's number and its traits by one small program, the engine. Same punk in, same voice out, on any computer, today or in ten years.
Step 1
Traits in
Every CryptoPunk is one of five types (Male, Female, Zombie, Ape or Alien) and wears up to seven attributes: hair, a beard, glasses, a pipe, an earring. That list comes straight from the original collection, and it is the only thing the engine reads besides the punk's number. No randomness from the outside, no clock, no server.
Step 2
Traits become a voice
Each trait nudges the voice somewhere, the way a face hints at how someone sounds.
The throat
Type sets the base pitch and the kind of voice. Males start near 112 Hz and females near 196 Hz. Zombies groan with a sub drone at 68 Hz, apes hoot and growl, and aliens don't use a throat at all: they chirp.
The delivery
A smile brightens the voice and makes it rise at the end. A frown drops it and adds rasp. Buck teeth add a lisp and a clown nose adds a nasal tone. Beards darken the voice and add a low hum underneath.
The extras
A pipe, cigarette or vape adds an inhale and an exhale. Shades add a glassy shimmer, VR adds a bit-crushed sweep, and eye shadow rings a note (C, D or E). Hats and helmets add a hit at the start.
What it says
Some traits decide the word: clown noses honk, buck teeth suck air, a smile laughs. Everyone else gets a grunt, a "huh", a "hey" or a sniff, picked from the punk's number.
The signature
Special punks carry a signature sound. Aliens send a transmission, apes rumble a demonic bass, zombies glitch. A trait worn by fewer than 150 punks earns one too: vinyl scratch, arc weld, music box.
The world
Every voice speaks from somewhere: a street, a subway, cassette hiss, CRT static, a club or an industrial hall. Welding goggles and pilot helmets always land in the industrial one.
Step 3
Synthesis
The engine builds the sound the way a throat does. A pulse train at the voice's pitch, with a little natural jitter, goes through formant filters: the resonances that make an "a" sound different from an "o". On top of that go breath, rasp and nasality, the word, the extras, the signature and the world.
The result is 2.6 seconds at 44.1 kHz, rendered straight to full-quality audio. There's no compression step anywhere.
Step 4
Same every time
Everything that looks random (which vowel, how it bends, which word) comes from a small random-number generator seeded with the punk's number. Run it twice and you get the same numbers, so you get the same voice. That is why a voice can belong to a punk: it isn't a recording that happened once, it's a result.
Step 5
The look
The art isn't generated from scratch either. It is the punk's own 24×24 pixels, every one in place, redrawn in white light. Skin turns white, black stays black and everything else becomes grey by its brightness. A glowing line runs around the outside, and three speech lines come out of the mouth, because now it talks.
Step 6
Locked on chain
The engine is one file. Its SHA-256 fingerprint is written into the NFT contract when the contract is created, and it can never be changed. Anyone can check that the engine this site runs is the one the collection promised:
engine hash
Change a single byte of the engine and the fingerprint no longer matches. At that point it would be a different collection, not an update to this one.