Sing-song: a speakable encoding for long numbers and keys

Sing-song is an encoding system for long numbers and keys that converts byte strings into pronounceable CV syllables. It uses a 64-syllable alphabet that maps each 6-bit value to a single syllable. The encoding is reversible, preserving byte length and leading zero bytes, and does not require external length metadata or punctuation.

The encoding is prefix-stable, meaning shared input prefixes produce shared syllable prefixes. It can be optionally represented with a variant suffix, providing alternative reversible representations of the same byte string. The encoding is deterministic, computationally simple, and can be spoken or transcribed without training.

The encoding alternates consonants and vowels, allowing for easy pronunciation. The final chunk may have fewer than six input bits, which are zero-filled and treated as canonical padding. The encoding preserves leading zero bytes and is self-sizing, meaning the length of the original byte string can be inferred from the encoding.

The decoder reconstructs the 6-bit chunks, infers the length, and returns the first 8·L bits. It rejects non-canonical syllable counts or non-zero padding. The encoding is designed to be easy to speak and transcribe, making it useful for generating deterministic usernames for Nostr npub keys.