Case · voice design & quality
How might someone shape a voice without touching a single setting?
Designing a voice by ear — for a voice platform.
Today shaping an AI voice usually means technical sliders — pitch, warmth, stability — which rewards expertise over instinct.
Why now as AI voices ship into products, media and games, the people picking them are rarely audio engineers.
Use cases Audiobooks, games, brand voices, accessibility, IVR.
The bigger play Picking by ear is really a low-threshold eval: every A/B choice is a preference judgement, and enough of them add up to a signal you can fine-tune and quality-check against — without anyone writing a rubric or reading a metric. That’s the harder problem underneath a friendly surface: making voice evaluation, fine-tuning and quality management feel intuitive, so non-engineers can steer quality and it still produces data a voice team can act on.
Interactive demo ↗The question
Most voice tools ask you to reason about sliders — pitch, pace, warmth. But people don't hear a voice as settings. They just know which one sounds right.
The approach
Skip the controls entirely. Play two takes of the same line and ask one thing: which do you prefer? Each choice steers the search toward the voice in your head.
The feature
Eight quick duels. A measurement field narrows with every pick until the voice settles. Nothing to configure — your ear does the tuning.
Function
The duel
Same recorded line, two candidates. Pick the better of the pair — no numbers, no sliders.
Function
The measurement field
An uncertainty band that tightens duel by duel — you can watch the voice converge.