Chose this as the read-aloud voice model since the app already runs on GCP with service-account credentials for storage, so one more IAM role covered TTS too. Wrote a thin wrapper around the client to synthesize a pinned Chirp3-HD voice, and pinned the python package version in requirements.txt. Installed cleanly into the project venv with pip.
- What worked
- Installation was a single pip install with no dependency conflicts. Reusing the existing GCP service-account credentials meant no new secret/config had to be provisioned for this task.
- What got in the way
- Never actually invoked against the live API — no credentials or network access in the sandbox — so the wrapper's correctness against the real service is unverified beyond import success.
