How Sound Determines Rebirth
Audio is first normalized to 16 kHz mono PCM, then cues such as pitch and pitch movement, spectral centroid and roll-off, bandwidth, flatness, zero-crossing rate, rhythm, event density, transients, and dynamic range are extracted. The system organizes these features into eight shared sound dimensions and determines whether the sound is closer to speech, a sustained tone, a whistle, an imitation, or another sound type.
Next, versioned rules place the sound into a deterministic 5×4 grid and select a base target from the current world’s character catalog. The current main fantasy worlds contain 400 base targets. Characters and worlds can expand, but the runtime does not use random numbers or hashes to arbitrarily pick an answer.
What Does “Reproducible” Mean?
The same standardized PCM, algorithm version, and catalog version should produce the same base result. Re-recording creates a different waveform even when the content is identical, so it may land on a neighboring coordinate. When versions are upgraded or the character catalog expands, the page should retain version information to help explain differences.
What Sounds Can You Try?
Natural speech, singing, humming, whistling, animal imitations, shouts, impacts, and ambient sounds can all enter the process. The system rejects only material that cannot be decoded, is nearly silent, is too short, or is severely clipped. Different sound types follow appropriate interpretation paths rather than all being treated as human speech.
A character’s gender tendency and age styling serve only the visual design of humanoid characters and are not judgments about the uploader. Rarity is based on the current frequency of results in the population and does not indicate character strength.
Recordings & Collection
Task audio exists only temporarily during analysis and is deleted afterward. Card snapshots, character results, and collection records may be retained long term, but they do not contain the original audio.