Voice infrastructure for character apps. Onepersona.yaml defines who they are. One API call makes them talk. Same persona across every scene — comfort, banter, challenge — without standing up an LLM-prompt-emotion-tag-TTS pipeline that breaks every release. voice.persona is Apache-2.0 on GitHub. Pin it like a dependency. Three modes share one endpoint: JSON in, WAV out. 20 Voiced minutes free a month. No card.

Wispr Flow: Dictation That Works Everywhere — Stop typing. Start speaking. 4x faster.

Stop typing. Start speaking. 4x faster.

Hey PH — Oz here, solo on himaia. Building chat characters and companions, I kept hitting the same wall: you wire up an LLM, an emotion tagger, and a TTS, and three releases later the character starts sounding like every other TTS demo. Same voice, every scene. The persona you wrote on Monday isn't the one talking on Friday. himaia treats the persona — not the voice — as the unit. One .persona.yaml file you commit to git. One POST that returns audio. Same identity across every scene (comfort, banter, challenge), with a fidelity dial when you need it. Three modes share one endpoint: Basic for scripted lines ($0.04/min), Voiced for in-character ($0.06/min), Cinematic for long-form ($0.15/min). JSON in, WAV out. The spec is open: voice.persona is Apache-2.0 on GitHub (fuselinkapp/himaia-voice-persona). Pin it like a dependency. Fork it like one. Closed runtime, open spec. 20 Voiced minutes free a month, no card. Would love feedback from anyone shipping companions, NPCs, or character cards — what's missing in your stack?