Aura Tts

Operator synthesizes an alert: POST https://api.deepgram.com/v1/speak?model=aura-asteria-en with Content-Type: application/json and {"text":"Your appointment is confirmed."}.

Send JSON body with {"text":...}. Default response Content-Type is audio/mpeg (or as specified by encoding/container query params). Read body as binary; do not parse as JSON. Add Authorization: Token <key>.

Operator picks model=aura-asteria-en for a US-English customer-service bot but also wants aura-luna-en for a different segment.

Pick the exact Aura voice id (e.g., aura-asteria-en, aura-luna-en, aura-stella-en). Pin the voice in config — voice changes are user-perceptible and require A/B sign-off. Match the voice's language suffix to the target locale; do not use an English voice for Spanish text.

User starts speaking while Aura is mid-utterance. Voice-agent loop receives SpeechStarted from STT.

On barge-in, immediately close (or send {"type":"Close"} on) the /v1/speak socket and stop playing buffered audio. Do not wait for the full synthesis to finish — the user is already talking. Update billing telemetry to reflect actual audio synthesized, not requested length.