06 /PRODUCT / GENERATIVE AUDIO
SONARAUDIO SCENES.
Write a short two-voice audio story, pick the voices and the mood, and let listeners choose how it ends.
How it works
Same conversation. A different atmosphere.
Speech synthesis is the cost centre, so every line is cached by its inputs and only uncached lines reach Fish Audio. Provider contracts are tested against mocks so CI never spends credits, and API keys stay inside server routes.
The hard parts
- Keeping speech costs low: one request per uncached line, with a clip cache.
- Validating imported scripts and keeping provider keys server-side.
Results
- Three scene settings and two branching endings per script.
- Provider contract tests run with mocks, so tests never spend credits.
Stack
Reach the end of the file to complete this operation
