Artificial Humanity builds deeply expressive, local-first speech synthesis โ€” the framework layer between text and genuinely dramatic spoken performance, running entirely on-device.

๐ŸŽญ Prosodia โ€” the on-device dramatic audiobook engine

Most text-to-speech reads. Prosodia performs. Every book is staged as a production: a Director (an on-device LLM) reads ahead and annotates each passage with emotional direction โ€” valence, arousal, tension; an Actor (neural TTS) performs those notes through a Rust synthesis core; a Stage coordinator keeps the audio gap-free. No cloud, no telemetry โ€” no page of your book leaves your hands.

๐ŸŽ™๏ธ Sonora โ€” training the voices

Our voice-actor training project. The first trained voice is live in the Sonora model registry โ€” fidelity-verified ONNX and LiteRT/TFLite artifacts, including a mobile-ready split-graph export. You can hear it in the Prosodia Space. Emotional directability (VAT conditioning) and a multi-voice casting grid are in active development.


๐Ÿ”— GitHub ยท ๐ŸŽง Hear the voice ยท ๐ŸŽ™๏ธ Sonora models