Hi jimmy,
Short answer: Vapi does not currently support real-time emotion recognition for assistant outputs. Our orchestration layer extracts emotion for internal use (see "Emotion detection" in How Vapi Works:
https://docs.vapi.ai/how-vapi-works) but there is no supported end-user feature that returns emotion data in the assistant/chat API or dashboard.
Notes and alternatives:
• No dashboard toggle or assistant setting exposes emotion recognition today.
• Some older/experimental voice models previously supported emotion controls for TTS (Cartesia older models), but newer models (e.g., sonic-2) do not — see related threads mentioning that limitation.
• If you need emotion/sentiment for post-call analysis, consider sending call audio or transcripts to a separate speech/emotion service (or an LLM) after the call.