1
Create an API-console project and keep its key in a server-side secret store. Choose between a real-time voice session, text-to-speech, or transcription according to the feature you need. Follow the current guide for audio formats, authentication, and the connection protocol. For a voice agent, set clear instructions and expose only the tools needed to complete the user’s request. Connect audio input and output, then test turn-taking, silence, interruptions, and recovery after a dropped connection. For batch speech tasks, inspect the returned text or audio before using it downstream. Review microphone permissions and explain recording behavior in the application. Check each service’s current billing unit and rate, since minutes, characters, and other usage measures are not interchangeable.