Voice Control And Prosody

Agent sets a speech speed control to an extreme out-of-range value to make audio 'very fast' for a skip feature.

Set the documented speed control within its supported range/enum. Clamp or reject out-of-range values client-side and surface a clear error rather than sending an invalid value and retrying. Prefer the documented control over post-hoc audio time-stretching that degrades quality.

Operator wants an 'excited' delivery and invents an undocumented emotion field rather than using the controls the API exposes.

Use only the documented expressivity/emotion controls and their supported values [REQUIRES-VERIFICATION on the exact control names and enums per current docs]. If a desired affect is not exposed, do not fabricate a field — adjust via supported controls/voice choice and flag the gap.

A brand name and a drug name are consistently mispronounced; the agent edits the transcript with ad-hoc misspellings to force pronunciation.

Use the documented pronunciation control (custom pronunciation / phoneme or dictionary mechanism) to fix specific terms deterministically, keeping the user-visible transcript clean. Maintain a reusable term→pronunciation map rather than corrupting the source text per occurrence.