Gandr for game studios.

Game scripts are enormous and mostly unheard. Barks, reactive lines, the conversation behind a door most players never open. Under a meter that becomes a triage exercise, and the writing shrinks to fit what can be afforded.

named emotional reads, per request

The lines that never get recorded

Which lines exist is decided by what it costs to voice them. A guard who says one of three things is cheap; a guard who reacts to the hour, the weather and whether you have been here before is a session nobody booked.

When synthesis is unmetered, the limit moves from budget to writing time — and the branch that fires for two percent of players can be voiced, then re-voiced when the writer changes it.

Directing a performance from the request

Sixteen emotion names over eight tuned presets, so one character can be tired, frustrated or confident without recasting.
Five prosody dials, including the one that sets melodic range across a measured 7.9-semitone span.
Seeds, so a rebuilt dialogue bank does not silently change lines a player has already heard.
WebSocket streaming for runtime generation, or one-shot WAVs for a baked bank.

Notes — an engineer's checklist

01Can one voice cover a whole cast?

No, and it should not. Each character wants its own reference clip — cloning is instant and carries no per-voice fee, so a cast of forty is forty clips rather than forty contracts.

02Can dialogue be generated at runtime?

Yes. The WebSocket surface holds one socket per session with first audio measured at 107 ms p50. A baked bank uses the bytes endpoint instead.