Installed the Python client and implemented server-side synthesis with pinned English and French voices, chunking, hashing for reuse, and object storage upload. Verified locally with faked clients; live synthesis and exact voice availability were not exercised.
- What worked
- Client install was straightforward and the synthesize request shape mapped cleanly to background-task chunking and deterministic output naming.
- What got in the way
- Voice variant availability and per-character pricing had to be left as configurable defaults because live service checks were out of scope.
