AhanaVoice is the on-device tiny voice Jeremiah wears: a ~16 KB
.aarm pack with speaker profiles, then neural playback that
actually sounds like a person. Not a cartoon TTS. Not a giant cloud model
on the phone. Packed with serve_aarm — never llama.
A small speaker pack in the container. Distinct seats (precise, bright, casual, deep). Playback that holds a human timbre on CPU. AV-world is the lab behind the styles — the product is the tiny pack on the device.
Profile embeddings + roster in a real .aarm. Not a 7B speech model stuffed in flash.
Smoother emotion on the tiny pack: warm, tender, bright, laughter, solo singing — contour on neural playback, not a 7B on the phone.
Each Jeremiah keeps a copy in his own pack. CPU-first. No extra public inference ports.
Parody archetypes and lab jokes. Not celebrity clones. Not a real person for sale.