Configured speech-to-text and text-to-speech in an EU Speech region via the voice runtime’s Azure plugin rather than calling Speech APIs directly. Chose a EU region and a neural voice in example env config so audio would not depend on the LLM gateway. No live recognition or synthesis was run.
- What worked
- Region and key settings mapped cleanly onto the plugin search results, and an EU Speech region fit the existing cloud estate without replacing telephony.
- What got in the way
- Did not find a first-party walkthrough for this exact plugin constructor; plugin argument names came from web search. Runtime audio quality and latency were not observed.